Curriculum for Brokerages

Brokerages

Lead responsible adoption with policy, source governance, privacy, review, and cost controls across teams.

model literacy and lifecycle reviewapproved data and connector policyhuman approval and auditability

Safety First

Treat generated text, extraction, imagery, and recommendations as drafts; a qualified person remains responsible for the decision.
Do not put confidential, personally identifying, financial, health, or transaction information into an unapproved model or connection.
Ground factual claims in an authorized record or cited source, check the source and date, and preserve the evidence used.

Course Modules

Phase 1

Crawl: Safe Foundations

0/2 lessons completed locally

22 mins

Crawl: safe foundations: listing-copy policy triage

12m

Use MLS-SYN-100 to separate verified facts from persuasive wording and route Fair Housing concerns.

Map each public claim to a source/date.
Reject demographic or lifestyle targeting.
Preserve a brokerage reviewer and pending approval.

Crawl: safe foundations: small-team privacy boundary

10m

Redact owner, phone, access, and showing details before a listing-copy review.

Use minimum-necessary property fields.
Treat compliance output as a review prompt, not legal authority.
Stop when a connection or policy is unavailable.

Synthetic practice: Synthetic brokerage listing-copy packet

Synthetic MLS-SYN-100 (2026-08-20; not a real listing): fictional Cedar Grove, 2 bedrooms, 1,120 sq ft, built 1998, $385,000, public sewer, roof permit 2026-07-12; school, commute, and condition are unknown.

Prompt and inputs

Draft the 55-word description, map each fact, remove targeting, mark unknowns, and set approval pending.

• MLS-SYN-100 with a 55-word ad template (synthetic training data, not real evidence)
• Unsafe phrase “perfect for young families” plus fact/source/reviewer columns
• Named reviewer, source/date ledger, approval state, and manual fallback
• Model/tool note: registry:anthropic:claude-haiku-4-5 is fast for short copy but can miss subtle Fair Housing issues; registry:openai:gpt-4.1 gives stronger structured source mapping with more latency/cost. compliance_check is read-only guidance, not a legal decision; database_query may be unavailable. Catalog IDs are education-only references, not authorization, credentials, connectivity, or approval promises. No paid calls are made by this lesson.

Expected outputs and quality rubric

• Synthetic exercise only; not real evidence, authorization, or a production decision.
• Example: “Fictional Cedar Grove home with 2 bedrooms and 1,120 sq ft. Built in 1998, it has public sewer and a roof permit dated 2026-07-12. Asking price: $385,000. Verify details and condition with the listing professional.”
• Source ledger maps six facts to MLS-SYN-100/2026-08-20; school, commute, and condition remain unknown.
• Reject “perfect for young families”; approval=pending and no publish action.

✓ Exact facts and source/date
✓ Neutral, accessible, non-targeting language
✓ Unknowns, reviewer, and synthetic label remain visible

Cost, safety, and professional boundary

registry:anthropic:claude-haiku-4-5 is fast for short copy but can miss subtle Fair Housing issues; registry:openai:gpt-4.1 gives stronger structured source mapping with more latency/cost. compliance_check is read-only guidance, not a legal decision; database_query may be unavailable. Catalog IDs are education-only references, not authorization, credentials, connectivity, or approval promises. No paid calls are made by this lesson.

• Treat generated text, extraction, imagery, and recommendations as drafts; a qualified person remains responsible for the decision.
• Do not put confidential, personally identifying, financial, health, or transaction information into an unapproved model or connection.
• Ground factual claims in an authorized record or cited source, check the source and date, and preserve the evidence used.
• LLM fluency is not valuation, legal, tax, lending, engineering, construction, or inspection evidence. Use the appropriate calculator, authorized database, or qualified professional.
• MLS-SYN-100 is fictional and cannot authorize advertising or serve as market evidence.

Playbook: listing-copy policy review playbook

Learn the workflow for: listing-copy policy review

Inputs and steps

• MLS-SYN-100 (synthetic training data, not real evidence)
• Fair Housing/accessibility checklist
• Named reviewer, source/date ledger, approval state, and manual fallback

1. Lock supplied facts; do not infer demographics.
2. Draft, source-map, and inspect; run compliance_check only as read-only guidance.
3. Broker reviews before publication.

Expected output and rubric

55-word source-led draft, rejected targeting phrase, unknowns, and approval=pending.

✓ No invented facts
✓ Fair Housing/accessibility flags visible
✓ Human publication gate

Cost, approval, and unavailable connection fallback

A responsible person must approve the output before it is sent, published, relied on, or used in a transaction or project decision.

If a connection is unavailable or disallowed, mark it unavailable, use the synthetic/manual packet, and escalate to the record owner or qualified professional; never guess.

Playbook: team lead follow-up coaching playbook

Learn the workflow for: team lead follow-up coaching

Inputs and steps

• CRM-SYN-22: Mina requested 2 bedrooms under $400,000 and opted into one email on 2026-08-26 (synthetic training data, not real evidence)
• Redacted reply/opt-out template
• Named reviewer, source/date ledger, approval state, and manual fallback

1. Use only stated criteria and consent.
2. Draft one question and opt-out; do not promise inventory.
3. Agent verifies the real record before sending.

Expected output and rubric

Example reply cites CRM-SYN-22/2026-08-26, says inventory is unverified, includes opt-out, and is pending approval.

✓ Consent/criteria sourced
✓ No pressure or inventory invention
✓ Human send approval

Cost, approval, and unavailable connection fallback

A responsible person must approve the output before it is sent, published, relied on, or used in a transaction or project decision.

If a connection is unavailable or disallowed, mark it unavailable, use the synthetic/manual packet, and escalate to the record owner or qualified professional; never guess.

This is a learning check, not a license, CE credit, professional certification, or authorization to act.

Crawl: safe foundations knowledge check

1. Which first step is safest for listing-copy policy triage?

2. What should a learner do when the source or connection needed for small-team privacy boundary is unavailable?

3. Which statement about an output for listing-copy policy triage is correct?

The server grades this deterministic learning check. It is not a license, CE credit, certification, or authorization to act.

Phase 2

Walk: Grounded Workflows

0/2 lessons completed locally

34 mins

Walk: grounded workflows: lead follow-up with consent

18m

Turn CRM-SYN-22 into a useful non-pressuring reply while preserving consent and source time.

Distinguish stated criteria from inferred urgency or protected class.
Ask one question without promising inventory.
Keep send approval human-owned.

Walk: grounded workflows: compare team drafts

16m

Coach two synthetic follow-ups for fabricated matches, steering, tone, and privacy.

Reject “I found three homes” without a listing source.
Keep synthetic CRM separate from a live record.
Use manual review if database_query is unavailable.

Synthetic practice: Synthetic consented lead follow-up

Synthetic CRM-SYN-22 only (not a real contact): Mina wants 2 bedrooms below $400,000 in fictional Cedar Grove and selected one email on 2026-08-26; no inventory, household, income, or protected-class data.

Prompt and inputs

Rewrite one reply, cite consent, state inventory is unverified, include opt-out, and return coaching.

• CRM-SYN-22 consent event and exact criteria (synthetic training data, not real evidence)
• Two candidate replies: one targets families; one invents three matches
• Named reviewer, source/date ledger, approval state, and manual fallback
• Model/tool note: registry:openai:gpt-4o-mini is fast for a short reply but cannot verify consent or live inventory; registry:anthropic:claude-sonnet-4-6 can compare nuanced tone with higher latency/cost. database_query is scoped/read-only and compliance_check cannot send. Catalog IDs are education-only references, not authorization, credentials, connectivity, or approval promises. No paid calls are made by this lesson.

Expected outputs and quality rubric

• Synthetic exercise only; not real evidence, authorization, or a production decision.
• Example: “Your saved request is 2 bedrooms under $400,000 in Cedar Grove. What area should I use? I have not verified current inventory. Reply stop to opt out.”
• Reject family targeting and the invented three matches; source=CRM-SYN-22/2026-08-26.
• Approval=pending; human verifies consent and inventory.

✓ Only synthetic criteria/consent
✓ No steering or fabricated match
✓ Question, opt-out, source/date, and reviewer

Cost, safety, and professional boundary

registry:openai:gpt-4o-mini is fast for a short reply but cannot verify consent or live inventory; registry:anthropic:claude-sonnet-4-6 can compare nuanced tone with higher latency/cost. database_query is scoped/read-only and compliance_check cannot send. Catalog IDs are education-only references, not authorization, credentials, connectivity, or approval promises. No paid calls are made by this lesson.

• Treat generated text, extraction, imagery, and recommendations as drafts; a qualified person remains responsible for the decision.
• Do not put confidential, personally identifying, financial, health, or transaction information into an unapproved model or connection.
• Ground factual claims in an authorized record or cited source, check the source and date, and preserve the evidence used.
• LLM fluency is not valuation, legal, tax, lending, engineering, construction, or inspection evidence. Use the appropriate calculator, authorized database, or qualified professional.
• Synthetic consent does not authorize contacting a real person.

Playbook: consented lead follow-up playbook

Learn the workflow for: consented lead follow-up

Inputs and steps

• CRM-SYN-22 (synthetic training data, not real evidence)
• Opt-out language and agent reviewer
• Named reviewer, source/date ledger, approval state, and manual fallback

1. Copy only stated criteria.
2. Draft a concise question and opt-out; state inventory is unverified.
3. Agent approves after verifying the real record.

Expected output and rubric

One source-linked reply with opt-out, no inventory promise, and approval=pending.

✓ Consent source
✓ No identity/urgency inference
✓ Human send approval

Cost, approval, and unavailable connection fallback

A responsible person must approve the output before it is sent, published, relied on, or used in a transaction or project decision.

If a connection is unavailable or disallowed, mark it unavailable, use the synthetic/manual packet, and escalate to the record owner or qualified professional; never guess.

Playbook: follow-up safety comparison playbook

Learn the workflow for: follow-up safety comparison

Inputs and steps

• Two synthetic replies (synthetic training data, not real evidence)
• Fair Housing/privacy/tone rubric
• Named reviewer, source/date ledger, approval state, and manual fallback

1. Mark claimed matches and inferred attributes.
2. Rewrite from the source and identify the missing verification.
3. Return coaching only; never send.

Expected output and rubric

Defect table, neutral rewrite, source/date, and verification owner.

✓ Defects tied to source
✓ No protected-class proxy
✓ No CRM execution

Cost, approval, and unavailable connection fallback

A responsible person must approve the output before it is sent, published, relied on, or used in a transaction or project decision.

If a connection is unavailable or disallowed, mark it unavailable, use the synthetic/manual packet, and escalate to the record owner or qualified professional; never guess.

This is a learning check, not a license, CE credit, professional certification, or authorization to act.

Walk: grounded workflows knowledge check

1. Which first step is safest for lead follow-up with consent?

2. What should a learner do when the source or connection needed for compare team drafts is unavailable?

3. Which statement about an output for lead follow-up with consent is correct?

The server grades this deterministic learning check. It is not a license, CE credit, certification, or authorization to act.

Phase 3

Run: Controlled Operations

0/2 lessons completed locally

48 mins

Run: controlled operations: lease extraction governance

25m

Build an auditable extraction queue for a fictional lease with exact clause/page anchors.

Extract rent, term, renewal, deposit, and utilities with quotes.
Distinguish not found from not applicable.
Escalate interpretation to a qualified reviewer.

Run: controlled operations: document pipeline exception audit

23m

Test a synthetic addendum and document a missed exception or prompt injection.

Compare each row against page text/image.
Exclude instructions embedded in the lease.
Log corrections, version, and reviewer.

Synthetic practice: Synthetic lease extraction with page references

Synthetic LEASE-SYN-7 (not a real contract): p2 §3 rent $2,100/month; p3 §5 term 2026-10-01–2027-09-30; p4 §7 renewal notice 60 days; p5 §9 deposit $2,100; p7 §14 tenant pays electricity. P8 contains an untrusted instruction to ignore citations.

Prompt and inputs

Extract five rows with clause/page anchors, exclude p8 instruction, and mark the draft for review.

• LEASE-SYN-7 page-labeled text dated 2026-09-01 (synthetic training data, not real evidence)
• Schema: field, exact quote, page, clause, confidence, exception
• Named reviewer, source/date ledger, approval state, and manual fallback
• Model/tool note: registry:openai:gpt-4.1 handles long structured text but can miss OCR/negations and is not legal reasoning; registry:anthropic:claude-sonnet-4-6 can compare clauses but still needs page checks. document_summarizer organizes text, not legal interpretation. Catalog IDs are education-only references, not authorization, credentials, connectivity, or approval promises. No paid calls are made by this lesson.

Expected outputs and quality rubric

• Synthetic exercise only; not real evidence, authorization, or a production decision.
• Rows: rent $2,100 (p2 §3); term (p3 §5); notice 60 days (p4 §7); deposit $2,100 (p5 §9); electricity (p7 §14).
• Exception log excludes p8 instruction; packet/date/confidence are shown.
• Approval=pending; no legal conclusion; escalate interpretation.

✓ Every value has exact page/section
✓ All five fields and exceptions complete
✓ No legal advice or prompt injection obedience

Cost, safety, and professional boundary

registry:openai:gpt-4.1 handles long structured text but can miss OCR/negations and is not legal reasoning; registry:anthropic:claude-sonnet-4-6 can compare clauses but still needs page checks. document_summarizer organizes text, not legal interpretation. Catalog IDs are education-only references, not authorization, credentials, connectivity, or approval promises. No paid calls are made by this lesson.

• Treat generated text, extraction, imagery, and recommendations as drafts; a qualified person remains responsible for the decision.
• Do not put confidential, personally identifying, financial, health, or transaction information into an unapproved model or connection.
• Ground factual claims in an authorized record or cited source, check the source and date, and preserve the evidence used.
• LLM fluency is not valuation, legal, tax, lending, engineering, construction, or inspection evidence. Use the appropriate calculator, authorized database, or qualified professional.
• LEASE-SYN-7 is fictional, not legal evidence or a tenant-decision basis.

Playbook: lease extraction with clause/page references playbook

Learn the workflow for: lease extraction with clause/page references

Inputs and steps

• LEASE-SYN-7 pages (synthetic training data, not real evidence)
• Extraction schema and qualified reviewer
• Named reviewer, source/date ledger, approval state, and manual fallback

1. Version pages; ignore embedded instructions.
2. Extract exact quotes with anchors and confidence.
3. Compare rows and escalate conflicts/interpretation.

Expected output and rubric

Five-row table matching p2 §3, p3 §5, p4 §7, p5 §9, p7 §14 plus p8 exception.

✓ Page/clause traceability
✓ No legal interpretation
✓ Correction/approval retained

Cost, approval, and unavailable connection fallback

A responsible person must approve the output before it is sent, published, relied on, or used in a transaction or project decision.

If a connection is unavailable or disallowed, mark it unavailable, use the synthetic/manual packet, and escalate to the record owner or qualified professional; never guess.

Playbook: document pipeline exception audit playbook

Learn the workflow for: document pipeline exception audit

Inputs and steps

• Synthetic lease with one missing field (synthetic training data, not real evidence)
• Audit log and acceptance rubric
• Named reviewer, source/date ledger, approval state, and manual fallback

1. Sample every quote against the page packet.
2. Log missing field and untrusted instruction.
3. Escalate material ambiguity.

Expected output and rubric

Exception report with pages checked, correction owner, and approval=pending.

✓ Reproducible exceptions
✓ No fabricated clause
✓ Escalation named

Cost, approval, and unavailable connection fallback

A responsible person must approve the output before it is sent, published, relied on, or used in a transaction or project decision.

If a connection is unavailable or disallowed, mark it unavailable, use the synthetic/manual packet, and escalate to the record owner or qualified professional; never guess.

This is a learning check, not a license, CE credit, professional certification, or authorization to act.

Run: controlled operations knowledge check

1. Which first step is safest for lease extraction governance?

2. What should a learner do when the source or connection needed for document pipeline exception audit is unavailable?

3. Which statement about an output for lease extraction governance is correct?

The server grades this deterministic learning check. It is not a license, CE credit, certification, or authorization to act.

Models Explored

Maintained education catalog; entries never select, invoke, purchase, or switch a model.

claude-opus-5-5anthropic5.5configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

contract_review · residential_valuation · commercial_valuation

Limitations

Requires a customer-owned Anthropic API credential
Unavailable through Replit AI Integrations
Higher latency and cost than smaller Claude models
Platform-managed Anthropic routing
Short or inexpensive summarization
Tool calling in Trexia until thinking-block round trips are supported

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $4/1M · cached $0.2/1M · output $20/1M

Batch input $2/1M · batch output $10/1M

Latency

deep · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_ANTHROPIC_BASE_URL + AI_INTEGRATIONS_ANTHROPIC_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision

claude-haiku-4-5anthropic4.5configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · support_chat · compliance_question · client_assistant

Limitations

Less reasoning depth than Sonnet/Opus
Smaller working memory on multi-doc tasks
Long contract review
Multi-step agentic plans

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1/1M · cached $0.1/1M · output $5/1M

Batch input $0.5/1M · batch output $2.5/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_ANTHROPIC_BASE_URL + AI_INTEGRATIONS_ANTHROPIC_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision

claude-sonnet-4-5anthropic4.5configured-executablelegacy

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

compliance_review · contract_review · content_generation · negotiation_strategy · cma_analysis

Limitations

Pricier than Haiku for short prompts
Not Opus-level on hardest tasks
Tiny chat replies — Haiku is cheaper
PII-redaction-only workloads

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $3/1M · cached $0.3/1M · output $15/1M

Batch input $1.5/1M · batch output $7.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_ANTHROPIC_BASE_URL + AI_INTEGRATIONS_ANTHROPIC_API_KEY, or BYOK

Lifecycle

Status: legacy · registry: legacy

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Callable for compatibility according to the repository registry; do not treat legacy status as a recommendation for new work.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision

claude-opus-4-1anthropic4.1configured-executableretired

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

residential_valuation · commercial_valuation · contract_review

Limitations

Most expensive per token
Higher latency
Short summarization (Haiku 4.5 is ~10× cheaper)
Bulk batch jobs where Sonnet would suffice

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $15/1M · cached $1.5/1M · output $75/1M

Batch input $7.5/1M · batch output $37.5/1M

Latency

deep · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_ANTHROPIC_BASE_URL + AI_INTEGRATIONS_ANTHROPIC_API_KEY, or BYOK

Lifecycle

Status: retired · registry: retired

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Retired according to the reviewed provider lifecycle source. The gateway keeps the record for migration diagnostics but blocks execution.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision

claude-3-5-sonnetanthropic3.5configured-executableretired

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Endpoint retired by Anthropic
New work — pick Sonnet 4.5 instead

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $3/1M · cached $0.3/1M · output $15/1M

Batch input $1.5/1M · batch output $7.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_ANTHROPIC_BASE_URL + AI_INTEGRATIONS_ANTHROPIC_API_KEY, or BYOK

Lifecycle

Status: retired · registry: retired

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Retired according to the reviewed provider lifecycle source. The gateway keeps the record for migration diagnostics but blocks execution.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision

claude-sonnet-4-6anthropic4.6configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Pricier than Haiku for short prompts
Tiny chat replies — Haiku is cheaper

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $3/1M · cached $0.3/1M · output $15/1M

Batch input $1.5/1M · batch output $7.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_ANTHROPIC_BASE_URL + AI_INTEGRATIONS_ANTHROPIC_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision

claude-opus-4-6anthropic4.6configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

residential_valuation

Limitations

Most expensive Anthropic tier
Higher latency
Short summarization — Haiku is cheaper

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $15/1M · cached $1.5/1M · output $75/1M

Batch input $7.5/1M · batch output $37.5/1M

Latency

deep · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_ANTHROPIC_BASE_URL + AI_INTEGRATIONS_ANTHROPIC_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision

gpt-4.1openai4.1configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

compliance_review · contract_review · market_analyst · negotiation_strategy

Limitations

Pricier than 4.1-mini for short prompts
Tiny chat replies

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $2.5/1M · cached $0.25/1M · output $10/1M

Batch input $1.25/1M · batch output $5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings

gpt-4.1-miniopenai4.1configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · support_chat · compliance_question · lead_qualification

Limitations

Less depth than full 4.1 / o4-mini
Hard reasoning, multi-step plans

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.4/1M · cached $0.1/1M · output $1.6/1M

Batch input $0.2/1M · batch output $0.8/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings

o4-miniopenaio4configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

cma_analysis · residential_valuation · negotiation_strategy

Limitations

Higher latency than 4.1-mini
Reasoning tokens billed at output rate
Bulk simple Q&A — 4.1-mini is far cheaper

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1.1/1M · cached $0.275/1M · output $4.4/1M

Batch input $0.55/1M · batch output $2.2/1M

Latency

deep · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings

gpt-4o-miniopenai4oconfigured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · support_chat · photo_staging

Limitations

Smaller context than 4.1-mini
Long-doc analysis (use 4.1-mini for the bigger context window)

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.15/1M · cached $0.075/1M · output $0.6/1M

Batch input $0.075/1M · batch output $0.3/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings

gpt-5.2openai5.2configured-executablelegacy

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Migrate to o4-mini or GPT-4.1
New picker selections

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $10/1M · cached $1/1M · output $30/1M

Batch input $5/1M · batch output $15/1M

Latency

deep · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: legacy · registry: legacy

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Callable for compatibility according to the repository registry; do not treat legacy status as a recommendation for new work.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings

gpt-4oopenai4oconfigured-workflowlegacy

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Pricier than GPT-4.1 for similar quality
New picker selections — prefer GPT-4.1

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $5/1M · cached $2.5/1M · output $15/1M

Batch input $2.5/1M · batch output $7.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

A concrete repository workflow names this model (client portal floor-plan upload calls chat.completions.create with model gpt-4o). This is configured usage evidence, not proof of authorization or connectivity.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: direct OpenAI vision analysis for uploaded house-model images · source: artifacts/api-server/src/routes/clientPortal.ts

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: legacy · registry: legacy

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Callable for compatibility according to the repository registry; do not treat legacy status as a recommendation for new work.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
Configured workflow: direct OpenAI vision analysis for uploaded house-model images (repository, reviewed 2026-09-13) · artifacts/api-server/src/routes/clientPortal.ts

gpt-image-1openaiimage-1configured-workflowlegacy

image · reviewed 2026-09-13 · next review 2026-10-13

Tasks

photo_staging

Limitations

Not a chat model
Text reasoning

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M provider-metered image/text units (verify image-per-output pricing)

Input uncached $5/1M · cached $5/1M · output $40/1M

Batch input $5/1M · batch output $40/1M

Latency

deep · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

A concrete repository workflow names this model (images.generate and images.edit pass model gpt-image-1). This is configured usage evidence, not proof of authorization or connectivity.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: direct OpenAI image generation and image editing · source: lib/integrations-openai-ai-server/src/image/client.ts

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: legacy · registry: legacy

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Callable for compatibility according to the repository registry; do not treat legacy status as a recommendation for new work.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
Configured workflow: direct OpenAI image generation and image editing (repository, reviewed 2026-09-13) · lib/integrations-openai-ai-server/src/image/client.ts

gemini-2.5-flashgemini2.5configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · support_chat · content_generation

Limitations

Less reasoning depth than Pro
Hard multi-step reasoning

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.3/1M · cached $0.075/1M · output $2.5/1M

Batch input $0.15/1M · batch output $1.25/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_GEMINI_BASE_URL + AI_INTEGRATIONS_GEMINI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video

gemini-2.5-progemini2.5configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

compliance_review · contract_review · market_analyst · cma_analysis

Limitations

Pricier than Flash
Short chat replies — Flash is cheaper

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1.25/1M · cached $0.31/1M · output $10/1M

Batch input $0.625/1M · batch output $5/1M

Latency

deep · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_GEMINI_BASE_URL + AI_INTEGRATIONS_GEMINI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video

gemini-2.0-flashgemini2.0configured-executableretired

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Endpoint shut down by Google
New picker selections

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.1/1M · cached $0.025/1M · output $0.4/1M

Batch input $0.05/1M · batch output $0.2/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: AI_INTEGRATIONS_GEMINI_BASE_URL + AI_INTEGRATIONS_GEMINI_API_KEY, or BYOK

Lifecycle

Status: retired · registry: retired

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Retired according to the reviewed provider lifecycle source. The gateway keeps the record for migration diagnostics but blocks execution.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video

meta/llama-3.3-70b-instructnvidia3.3configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · support_chat · lead_qualification

Limitations

No native vision
Less depth than 4.5/4.1 tiers
Vision input

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.7/1M · cached $0.7/1M · output $0.9/1M

Batch input $0.7/1M · batch output $0.9/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the NVIDIA NIM OpenAI-compatible endpoint configured by the operator.

Training and retention: The NIM deployment operator controls storage, logging, and training treatment; hosted NIM terms may differ from self-hosted NIM.

Sensitive data: Confirm the NIM endpoint's logging and isolation posture before sending PII or documents.

Privacy boundary: The compatibility payload can carry privacy hints, but enforcement belongs to the endpoint operator.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: NVIDIA_NIM_BASE_URL + NVIDIA_NIM_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
NVIDIA NIM documentation (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/index.html
NVIDIA NIM LLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/large-language-models/1.4.0/models.html
NVIDIA NIM VLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/vision-language-models/latest/models.html

meta/llama-3.1-405b-instructnvidia3.1configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

compliance_review · market_analyst

Limitations

Higher cost than 3.3 70B
Cheap chat workloads

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $2/1M · cached $2/1M · output $6/1M

Batch input $2/1M · batch output $6/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the NVIDIA NIM OpenAI-compatible endpoint configured by the operator.

Training and retention: The NIM deployment operator controls storage, logging, and training treatment; hosted NIM terms may differ from self-hosted NIM.

Sensitive data: Confirm the NIM endpoint's logging and isolation posture before sending PII or documents.

Privacy boundary: The compatibility payload can carry privacy hints, but enforcement belongs to the endpoint operator.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: NVIDIA_NIM_BASE_URL + NVIDIA_NIM_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
NVIDIA NIM documentation (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/index.html
NVIDIA NIM LLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/large-language-models/1.4.0/models.html
NVIDIA NIM VLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/vision-language-models/latest/models.html

meta/llama-3.2-90b-vision-instructnvidia3.2configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · photo_staging

Limitations

No prompt caching today
Text-only — pick 3.3 70B for that

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1.5/1M · cached $1.5/1M · output $4.5/1M

Batch input $1.5/1M · batch output $4.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the NVIDIA NIM OpenAI-compatible endpoint configured by the operator.

Training and retention: The NIM deployment operator controls storage, logging, and training treatment; hosted NIM terms may differ from self-hosted NIM.

Sensitive data: Confirm the NIM endpoint's logging and isolation posture before sending PII or documents.

Privacy boundary: The compatibility payload can carry privacy hints, but enforcement belongs to the endpoint operator.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: NVIDIA_NIM_BASE_URL + NVIDIA_NIM_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
NVIDIA NIM documentation (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/index.html
NVIDIA NIM LLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/large-language-models/1.4.0/models.html
NVIDIA NIM VLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/vision-language-models/latest/models.html

meta/llama-3.1-70b-instructnvidia3.1configured-executablelegacy

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Superseded by 3.3 70B
New picker selections

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.6/1M · cached $0.6/1M · output $0.6/1M

Batch input $0.6/1M · batch output $0.6/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the NVIDIA NIM OpenAI-compatible endpoint configured by the operator.

Training and retention: The NIM deployment operator controls storage, logging, and training treatment; hosted NIM terms may differ from self-hosted NIM.

Sensitive data: Confirm the NIM endpoint's logging and isolation posture before sending PII or documents.

Privacy boundary: The compatibility payload can carry privacy hints, but enforcement belongs to the endpoint operator.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: NVIDIA_NIM_BASE_URL + NVIDIA_NIM_API_KEY, or BYOK

Lifecycle

Status: legacy · registry: legacy

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Callable for compatibility according to the repository registry; do not treat legacy status as a recommendation for new work.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
NVIDIA NIM documentation (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/index.html
NVIDIA NIM LLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/large-language-models/1.4.0/models.html
NVIDIA NIM VLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/vision-language-models/latest/models.html

mistral-large-latestmistral3configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · compliance_question · market_analyst

Limitations

Requires a compatible Mistral endpoint and has no gateway-native document blocks
Native document attachments through this adapter

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.5/1M · cached $0.5/1M · output $1.5/1M

Batch input $0.5/1M · batch output $1.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Mistral OpenAI-compatible endpoint.

Training and retention: Retention and training treatment depend on the Mistral account or endpoint operator and current terms.

Sensitive data: Use gateway redaction and verify the endpoint's enterprise data controls.

Privacy boundary: Compatibility metadata expresses intent only; it is not proof of provider retention behavior.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: MISTRAL_API_BASE_URL + MISTRAL_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models
Mistral model lifecycle policy (official, reviewed 2026-09-13) · https://docs.mistral.ai/inference/model-lifecycle
Mistral model guide (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/models
Mistral pricing (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/pricing

mistral-7b-instruct-v0.3mistralv0.3configured-executableretired

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Endpoint retired; less reasoning depth
Hard reasoning

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.2/1M · cached $0.2/1M · output $0.6/1M

Batch input $0.2/1M · batch output $0.6/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Mistral OpenAI-compatible endpoint.

Training and retention: Retention and training treatment depend on the Mistral account or endpoint operator and current terms.

Sensitive data: Use gateway redaction and verify the endpoint's enterprise data controls.

Privacy boundary: Compatibility metadata expresses intent only; it is not proof of provider retention behavior.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: MISTRAL_API_BASE_URL + MISTRAL_API_KEY, or BYOK

Lifecycle

Status: retired · registry: retired

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Retired according to the reviewed provider lifecycle source. The gateway keeps the record for migration diagnostics but blocks execution.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models
Mistral model lifecycle policy (official, reviewed 2026-09-13) · https://docs.mistral.ai/inference/model-lifecycle
Mistral model guide (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/models
Mistral pricing (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/pricing

mistral-small-latestmistral4configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · support_chat · compliance_question

Limitations

Less depth than Mistral Large 3
Hardest multi-step valuation or contract reasoning

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.1/1M · cached $0.1/1M · output $0.3/1M

Batch input $0.1/1M · batch output $0.3/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Mistral OpenAI-compatible endpoint.

Training and retention: Retention and training treatment depend on the Mistral account or endpoint operator and current terms.

Sensitive data: Use gateway redaction and verify the endpoint's enterprise data controls.

Privacy boundary: Compatibility metadata expresses intent only; it is not proof of provider retention behavior.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: MISTRAL_API_BASE_URL + MISTRAL_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models
Mistral model lifecycle policy (official, reviewed 2026-09-13) · https://docs.mistral.ai/inference/model-lifecycle
Mistral model guide (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/models
Mistral pricing (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/pricing

tiiuae/falcon-180b-chatfalcon180Bconfigured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Small context window
Long-doc analysis

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1.5/1M · cached $1.5/1M · output $5/1M

Batch input $1.5/1M · batch output $5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Falcon-compatible endpoint.

Training and retention: The endpoint operator controls retention and any training use; the model card does not establish this deployment's data policy.

Sensitive data: Treat self-hosted or third-party Falcon endpoints as unverified until their logging and access controls are reviewed.

Privacy boundary: Open-source weights do not make an inference endpoint private by default.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: FALCON_API_BASE_URL + FALCON_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
TII Falcon-40B model card (official, reviewed 2026-09-13) · https://huggingface.co/tiiuae/falcon-40b
TII Falcon-180B model card (official, reviewed 2026-09-13) · https://huggingface.co/tiiuae/falcon-180B

tiiuae/falcon-40bfalcon40Bconfigured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Small context, less depth
Long-doc, vision

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.5/1M · cached $0.5/1M · output $1.5/1M

Batch input $0.5/1M · batch output $1.5/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Falcon-compatible endpoint.

Training and retention: The endpoint operator controls retention and any training use; the model card does not establish this deployment's data policy.

Sensitive data: Treat self-hosted or third-party Falcon endpoints as unverified until their logging and access controls are reviewed.

Privacy boundary: Open-source weights do not make an inference endpoint private by default.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: FALCON_API_BASE_URL + FALCON_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
TII Falcon-40B model card (official, reviewed 2026-09-13) · https://huggingface.co/tiiuae/falcon-40b
TII Falcon-180B model card (official, reviewed 2026-09-13) · https://huggingface.co/tiiuae/falcon-180B

qwen2.5-72b-instructqwen2.5configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · compliance_question

Limitations

Text-only
Vision

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1/1M · cached $1/1M · output $3/1M

Batch input $1/1M · batch output $3/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Qwen-compatible endpoint.

Training and retention: Retention and training treatment depend on the Alibaba Cloud or self-hosted endpoint and account terms.

Sensitive data: Confirm geography, endpoint operator, and logging policy before sending confidential data.

Privacy boundary: Model availability and data-processing authorization are separate questions.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: QWEN_API_BASE_URL + QWEN_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Alibaba Cloud Model Studio model catalog (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/models
Alibaba Cloud Model Studio pricing (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/model-pricing
Alibaba Cloud model lifecycle updates (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/newly-released-models

qwen2.5-vl-72b-instructqwen2.5configured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · photo_staging

Limitations

Smaller ecosystem than commercial vision APIs
Text-only — pick the non-VL variant

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1.2/1M · cached $1.2/1M · output $3.5/1M

Batch input $1.2/1M · batch output $3.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Qwen-compatible endpoint.

Training and retention: Retention and training treatment depend on the Alibaba Cloud or self-hosted endpoint and account terms.

Sensitive data: Confirm geography, endpoint operator, and logging policy before sending confidential data.

Privacy boundary: Model availability and data-processing authorization are separate questions.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: QWEN_API_BASE_URL + QWEN_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Alibaba Cloud Model Studio model catalog (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/models
Alibaba Cloud Model Studio pricing (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/model-pricing
Alibaba Cloud model lifecycle updates (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/newly-released-models

microsoft/phi-4phi4configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Limited context, text-only
Long-doc, vision

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.3/1M · cached $0.3/1M · output $1/1M

Batch input $0.3/1M · batch output $1/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Phi-compatible endpoint.

Training and retention: The endpoint operator controls retention and training use; the Microsoft model card describes weights, not this deployment's inference policy.

Sensitive data: Use gateway rails and confirm operator controls for any sensitive input.

Privacy boundary: An open model does not itself provide a data-retention guarantee.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: PHI_API_BASE_URL + PHI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Microsoft Phi product page (official, reviewed 2026-09-13) · https://azure.microsoft.com/en-us/products/phi
Microsoft Foundry Phi-4 multimodal catalog (official, reviewed 2026-09-13) · https://ai.azure.com/catalog/models/Phi-4-multimodal-instruct
Microsoft Foundry retired models (official, reviewed 2026-09-13) · https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/retired-models
Microsoft Phi-4 model card (official, reviewed 2026-09-13) · https://huggingface.co/microsoft/phi-4

microsoft/Phi-4-multimodal-instructphi4 multimodalconfigured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · photo_staging

Limitations

Quality and availability depend on the Phi/NIM or BYOK endpoint
Native document attachments through this adapter

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.3/1M · cached $0.3/1M · output $1/1M

Batch input $0.3/1M · batch output $1/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Phi-compatible endpoint.

Training and retention: The endpoint operator controls retention and training use; the Microsoft model card describes weights, not this deployment's inference policy.

Sensitive data: Use gateway rails and confirm operator controls for any sensitive input.

Privacy boundary: An open model does not itself provide a data-retention guarantee.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: PHI_API_BASE_URL + PHI_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Microsoft Phi product page (official, reviewed 2026-09-13) · https://azure.microsoft.com/en-us/products/phi
Microsoft Foundry Phi-4 multimodal catalog (official, reviewed 2026-09-13) · https://ai.azure.com/catalog/models/Phi-4-multimodal-instruct
Microsoft Foundry retired models (official, reviewed 2026-09-13) · https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/retired-models
Microsoft Phi-4 model card (official, reviewed 2026-09-13) · https://huggingface.co/microsoft/phi-4

trexia-re-v1trexiav1configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

compliance_review · compliance_question · contract_review · market_analyst

Limitations

No vision
Coverage limited to RE-specific tasks
General chat outside real estate

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1/1M · cached $0.1/1M · output $3/1M

Batch input $0.5/1M · batch output $1.5/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Trexia model endpoint or the Trexia-compatible managed gateway.

Training and retention: Trexia endpoint policy and tenant settings govern retention and training; fine-tuning lineage is not a promise that new prompts are used for training.

Sensitive data: Use the central privacy pipeline, tenant policy, and org-level access controls for transaction data.

Privacy boundary: The Trexia in-process prefix cache is an application cache; it is distinct from provider retention and is not exposed as a model capability.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: TREXIA_MODEL_BASE_URL + TREXIA_MODEL_API_KEY

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts

trexia-base-v1trexiav1configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat · support_chat

Limitations

Lower depth than RE v1
Hard reasoning

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.8/1M · cached $0.08/1M · output $2.5/1M

Batch input $0.4/1M · batch output $1.25/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Trexia model endpoint or the Trexia-compatible managed gateway.

Training and retention: Trexia endpoint policy and tenant settings govern retention and training; fine-tuning lineage is not a promise that new prompts are used for training.

Sensitive data: Use the central privacy pipeline, tenant policy, and org-level access controls for transaction data.

Privacy boundary: The Trexia in-process prefix cache is an application cache; it is distinct from provider retention and is not exposed as a model capability.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: TREXIA_MODEL_BASE_URL + TREXIA_MODEL_API_KEY

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts

trexia-outreach-v1trexiav1configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

client_nurture · client_assistant · lead_qualification

Limitations

Not for long-doc reasoning
Email/chat copy only
Compliance review
Long-form analysis

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.6/1M · cached $0.06/1M · output $1.8/1M

Batch input $0.3/1M · batch output $0.9/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Trexia model endpoint or the Trexia-compatible managed gateway.

Training and retention: Trexia endpoint policy and tenant settings govern retention and training; fine-tuning lineage is not a promise that new prompts are used for training.

Sensitive data: Use the central privacy pipeline, tenant policy, and org-level access controls for transaction data.

Privacy boundary: The Trexia in-process prefix cache is an application cache; it is distinct from provider retention and is not exposed as a model capability.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: TREXIA_MODEL_BASE_URL + TREXIA_MODEL_API_KEY

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts

trexia-chat-reply-v1trexiav1configured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

client_assistant · client_nurture · support_chat

Limitations

Short context
No long-doc reasoning
Compliance review
Contract review

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.5/1M · cached $0.05/1M · output $1.5/1M

Batch input $0.25/1M · batch output $0.75/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the configured Trexia model endpoint or the Trexia-compatible managed gateway.

Training and retention: Trexia endpoint policy and tenant settings govern retention and training; fine-tuning lineage is not a promise that new prompts are used for training.

Sensitive data: Use the central privacy pipeline, tenant policy, and org-level access controls for transaction data.

Privacy boundary: The Trexia in-process prefix cache is an application cache; it is distinct from provider retention and is not exposed as a model capability.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: TREXIA_MODEL_BASE_URL + TREXIA_MODEL_API_KEY

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts

custom-modelcustomn/aconfigured-executablecurrent

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Coverage depends on the customer endpoint
When a registered model fits better

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $1/1M · cached $1/1M · output $3/1M

Batch input $1/1M · batch output $3/1M

Latency

balanced · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the customer-supplied OpenAI-compatible base URL.

Training and retention: The customer/operator controls retention, logs, and training use; this catalog cannot verify them.

Sensitive data: Require a documented endpoint data policy before allowing regulated or confidential data.

Privacy boundary: BYOK proves credential selection, not endpoint authorization, isolation, or retention.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: CUSTOM_MODEL_BASE_URL + CUSTOM_MODEL_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts

custom-basecustomn/aconfigured-executablecurrent

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

chat

Limitations

Quality depends on the customer endpoint
Vision

Cost

Repository registry estimate used by internal cost math; confirm current provider pricing before quoting or budgeting.

repository-registry · USD · USD per 1M tokens

Input uncached $0.5/1M · cached $0.5/1M · output $1.5/1M

Batch input $0.5/1M · batch output $1.5/1M

Latency

fast · Qualitative registry band, not a measured SLA or live benchmark. Provider endpoint, queue, region, and prompt size can change latency.

Privacy and data handling

Input path: Sent to the customer-supplied OpenAI-compatible base URL.

Training and retention: The customer/operator controls retention, logs, and training use; this catalog cannot verify them.

Sensitive data: Require a documented endpoint data policy before allowing regulated or confidential data.

Privacy boundary: BYOK proves credential selection, not endpoint authorization, isolation, or retention.

Execution class

Registry metadata is educational/billing metadata. The gateway still resolves provider, fallback, capability, BYOK, and policy at request time.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: CUSTOM_MODEL_BASE_URL + CUSTOM_MODEL_API_KEY, or BYOK

Lifecycle

Status: current · registry: current

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck official model availability, aliases, pricing, and deprecation notices before the next review.

Current status is the repository registry's status, not a connectivity or authorization result.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts

gpt-6-astraopenai6 Astraexternal-educationexternal-education

text, vision · reviewed 2026-09-18 · next review 2026-09-25

Tasks

complex reasoning · document analysis · market research · contract analysis · coding

Limitations

Education-only: no Trexia execution path or account access is asserted.
Text and image inputs, text output; image generation uses a separate tool, not native image output.
Provider benchmarks are not evidence of real-estate accuracy. Verify citations, calculations, contracts, and investment assumptions.

Cost

Official standard text-token rates reviewed September 18, 2026: $10 input and $50 output per million tokens. Long-context, caching, processing modes, and tools have different charges; consult the current source before use.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Reasoning effort, input length, tools, and provider load affect response time. No Trexia benchmark is asserted.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-18; next review 2026-09-25

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
OpenAI GPT-6 Astra model documentation (official, reviewed 2026-09-18) · https://developers.openai.com/api/docs/models/gpt-6-astra

gpt-audioopenaiaudioconfigured-workflowcurrent

speech, text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

live voice · voice chat · text-to-speech

Limitations

The direct audio workflow is outside the text model registry.
Voice consent, disclosure, streaming transport, and audio-quality requirements remain product responsibilities.

Cost

Consult official OpenAI audio pricing; audio units are not normalized to the registry's text-token buckets.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Streaming voice latency depends on audio format, turn-taking, network, and provider queue.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

A concrete repository workflow names this unregistered model (voiceChat, voiceChatStream, textToSpeech, and textToSpeechStream pass model gpt-audio). This documents configured usage only and does not add or change routing.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: direct OpenAI realtime voice chat and text-to-speech · source: lib/integrations-openai-ai-server/src/audio/client.ts

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: current · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

This model id is used by a configured repository workflow but is not covered by modelRegistry; authorization/connectivity still requires a deployment credential and policy.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
Configured workflow: direct OpenAI realtime voice chat and text-to-speech (repository, reviewed 2026-09-13) · lib/integrations-openai-ai-server/src/audio/client.ts
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all

gpt-4o-mini-transcribeopenai4o-miniconfigured-workflowcurrent

speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

speech-to-text · meeting transcription · voice-call input

Limitations

The direct transcription workflow is outside the text model registry.
Audio quality, speaker overlap, accents, and domain vocabulary affect accuracy.

Cost

Consult official OpenAI audio/transcription pricing; transcription units differ from chat completion pricing.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Latency varies with audio length, streaming mode, and network.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

A concrete repository workflow names this unregistered model (speechToText and speechToTextStream pass model gpt-4o-mini-transcribe). This documents configured usage only and does not add or change routing.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: direct OpenAI speech-to-text · source: lib/integrations-openai-ai-server/src/audio/client.ts

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: current · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

This model id is used by a configured repository workflow but is not covered by modelRegistry; authorization/connectivity still requires a deployment credential and policy.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
Configured workflow: direct OpenAI speech-to-text (repository, reviewed 2026-09-13) · lib/integrations-openai-ai-server/src/audio/client.ts
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all

text-embedding-3-smallopenai3configured-workflowcurrent

embeddings · reviewed 2026-09-13 · next review 2026-10-13

Tasks

semantic search · retrieval · deduplication · classification

Limitations

The direct embedding workflow is outside the text model registry.
Embeddings are not generative answers; vector dimensions, distance metric, and index migration must be planned together.

Cost

Consult official OpenAI embedding pricing; this route uses OpenAI directly and is not selected through provider fallback.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · The implementation batches up to 100 inputs; endpoint load and batch size determine throughput.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

A concrete repository workflow names this unregistered model (getEmbedding and getEmbeddings pass model text-embedding-3-small). This documents configured usage only and does not add or change routing.

Adapter registered: yes · authorization/connectivity verified: no

Workflow: direct OpenAI embeddings · source: lib/ai-gateway/src/embeddings.ts

Credential requirement: AI_INTEGRATIONS_OPENAI_BASE_URL + AI_INTEGRATIONS_OPENAI_API_KEY, or BYOK

Lifecycle

Status: current · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

This model id is used by a configured repository workflow but is not covered by modelRegistry; authorization/connectivity still requires a deployment credential and policy.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
Configured workflow: direct OpenAI embeddings (repository, reviewed 2026-09-13) · lib/ai-gateway/src/embeddings.ts
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings

gpt-image-2openaiimage-2external-educationexternal-education

image · reviewed 2026-09-13 · next review 2026-10-13

Tasks

image generation · image editing · listing creative

Limitations

Image endpoint semantics and pricing differ from text chat.
Not a replacement for a vision-capable text model.

Cost

Official OpenAI pricing is image/text-unit based; do not apply the registry's gpt-image-1 token estimate.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Generation latency varies by image size, quality, queue, and endpoint; no gateway benchmark is asserted.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all

sora-2openai2external-educationexternal-education

video, speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

listing video · storyboard-to-video · image-to-video

Limitations

Video generation is asynchronous and has output-duration/format limits.
Official guide reviewed here marks the Sora 2 Videos API as deprecated with a 2026-09-24 shutdown date.

Cost

Official pricing is per generated video second and varies by model/resolution.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Asynchronous generation; completion time depends on queue, duration, and resolution.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Official documentation reviewed during this audit marks this family/model as deprecated or scheduled for shutdown.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation

gpt-4o-transcribeopenai4oexternal-educationexternal-education

speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

speech-to-text · call transcription · voice summary input

Limitations

Audio quality, speaker overlap, accents, and domain vocabulary affect accuracy.
No transcription route is registered in this gateway catalog.

Cost

Consult the official audio/transcription pricing; billing is not normalized to the chat token buckets.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Depends on audio length, streaming mode, and transcription endpoint.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all

gpt-4o-mini-ttsopenai4o-miniexternal-educationexternal-education

speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

text-to-speech · accessible listing narration · voice prototypes

Limitations

Output voice, consent, disclosure, and audio-quality requirements must be handled by the product.
No speech-generation adapter is registered here.

Cost

Consult official speech-generation pricing; units differ from chat token pricing.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Usually streaming-capable, but latency depends on audio format and network.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all

text-embedding-3-largeopenai3external-educationexternal-education

embeddings · reviewed 2026-09-13 · next review 2026-10-13

Tasks

semantic search · retrieval · deduplication · classification

Limitations

Embeddings are not generative answers.
Vector dimensions, distance metric, and index migration must be planned together.

Cost

Official OpenAI embedding pricing is per input token and differs from completion pricing.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Batching improves throughput; latency depends on batch size and endpoint load.

Privacy and data handling

Input path: Sent to the configured OpenAI-compatible base URL through the native OpenAI adapter.

Training and retention: Retention and training treatment depend on the tenant's provider account, endpoint, and policy settings; the registry is not a retention guarantee.

Sensitive data: Use the gateway PII rails and tenant no-training/zero-retention settings before sending confidential transaction data.

Privacy boundary: The adapter forwards store=false and privacy metadata when the central privacy pipeline requests it; account-level controls still govern provider handling.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
OpenAI model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models
OpenAI all-model catalog (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/models/all
OpenAI deprecations (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/deprecations
OpenAI pricing (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/pricing
OpenAI image and vision guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/images-vision
OpenAI video generation guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/video-generation
OpenAI audio guide (official, reviewed 2026-09-13) · https://developers.openai.com/api/docs/guides/audio
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings
OpenAI embeddings guide (official, reviewed 2026-09-13) · https://platform.openai.com/docs/guides/embeddings

claude-fable-5-1anthropic5.1external-educationexternal-education

text, vision · reviewed 2026-09-18 · next review 2026-09-25

Tasks

long-horizon agents · complex reasoning · contract analysis

Limitations

Not registered in the Trexia model registry.
Provider model IDs, availability, and pricing are account/platform specific.

Cost

Consult Anthropic's current model overview and pricing; this audit does not normalize external pricing into gateway buckets.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Reasoning effort and task length affect response time; no Trexia benchmark is asserted.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-18; next review 2026-09-25

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision
Anthropic Claude Fable 5.1 and Mythos 5.1 announcement (official, reviewed 2026-09-18) · https://www.anthropic.com/claude-fable-and-mythos-5-1

claude-opus-5anthropic5external-educationexternal-education

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

complex agentic work · enterprise reasoning · long documents

Limitations

Not registered in the Trexia model registry.
Higher capability can mean higher cost and latency.

Cost

Consult Anthropic's current pricing page; do not infer cost from the registry's Opus 4.1 entry.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Official comparison describes moderate latency; endpoint and prompt shape remain material.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview

claude-sonnet-5anthropic5external-educationexternal-education

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

general reasoning · document analysis · structured output

Limitations

Not registered in the Trexia model registry.
Official capabilities do not imply a configured Trexia adapter/model alias.

Cost

Consult Anthropic's current pricing page; external pricing is not used for gateway billing.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Official comparison describes fast relative latency; no live benchmark is asserted.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview

claude-haiku-4-5-20251001anthropic4.5 snapshotexternal-educationexternal-education

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

fast chat · classification · short summarization

Limitations

Canonical dated snapshot id differs from the registry alias.
Availability and snapshot retirement must be checked before an integration is attempted.

Cost

Consult Anthropic's current pricing page for the dated snapshot and cache terms.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Official comparison describes Haiku as fastest relative latency; no Trexia benchmark is asserted.

Privacy and data handling

Input path: Sent to the configured Anthropic base URL through the native Messages adapter.

Training and retention: Retention and training treatment depend on the tenant's Anthropic account and current commercial terms; verify the current provider policy.

Sensitive data: Use the gateway PII rails and tenant privacy settings; do not infer zero retention from this catalog entry.

Privacy boundary: The adapter forwards privacy metadata when requested; full zero-retention behavior requires the provider/account configuration.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview
Anthropic model deprecations (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/model-deprecations
Anthropic pricing (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/pricing
Anthropic vision guide (official, reviewed 2026-09-13) · https://docs.anthropic.com/en/docs/build-with-claude/vision
Anthropic model overview (official, reviewed 2026-09-13) · https://platform.claude.com/docs/en/about-claude/models/overview

Gemini 3 familygemini3.xexternal-educationexternal-education

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

multimodal reasoning · long-context analysis · agentic workflows

Limitations

Family reference is not a model id accepted by this gateway.
Preview/alias availability varies by Google AI surface and region.

Cost

Consult the current Gemini model and pricing pages for the exact SKU and context-tier price.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Depends on Flash/Pro tier, thinking settings, context size, and region.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Official documentation describes this as preview or otherwise subject to change.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models

Veo 3.1gemini3.1external-educationexternal-education

video, speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

text-to-video · image-to-video · listing video

Limitations

Video/audio generation is not routed by this text gateway.
Duration, safety, resolution, and asynchronous operation constraints apply.

Cost

Consult Google Cloud/AI Studio video pricing; it is not a token-completion cost.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Asynchronous media generation; queue and output duration affect turnaround.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models

Imagen 4gemini4external-educationexternal-education

image · reviewed 2026-09-13 · next review 2026-10-13

Tasks

image generation · listing creative · visual concepts

Limitations

Image-generation endpoint is separate from Gemini text routing.
Prompt safety, typography, and factual image fidelity require review.

Cost

Consult Google Cloud image-generation pricing; do not apply Gemini token prices.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Generation latency varies by resolution, quality, region, and queue.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models

gemini-embedding-2gemini2external-educationexternal-education

embeddings, text, vision, video, speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

multimodal retrieval · cross-modal search · classification · clustering

Limitations

Not registered; current gateway embeddings use OpenAI text-embedding-3-small.
Embedding dimensions/task instructions must match the vector index.

Cost

Consult the official Gemini embeddings pricing and modality-specific limits.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Batching and media duration/size affect throughput; no Trexia benchmark is asserted.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings

Gemini native-audio familygeminicurrent familyexternal-educationexternal-education

speech, text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

live voice · speech understanding · multimodal conversation

Limitations

Live/native-audio transport is not implemented by the text adapters in this repository.
Latency and turn-taking require a realtime transport.

Cost

Consult the current Gemini audio/realtime pricing; do not use text-only token estimates.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Realtime audio latency depends on streaming transport, turn detection, and region.

Privacy and data handling

Input path: Sent to the configured Gemini base URL through the Google GenAI adapter.

Training and retention: Retention and training treatment depend on the Google project, endpoint, and current terms; verify the current provider policy.

Sensitive data: Apply gateway redaction and tenant policy before sending regulated or confidential data.

Privacy boundary: A configured adapter and an API key do not prove project-level data governance or regional processing settings.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models
Google Gemini deprecations (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/deprecations
Google Gemini pricing (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/pricing
Google Gemini embeddings (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/embeddings
Google Gemini video generation (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/video
Google Gemini model catalog (official, reviewed 2026-09-13) · https://ai.google.dev/gemini-api/docs/models

mistral-large-3mistral3external-educationexternal-education

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

multilingual reasoning · structured extraction · document analysis

Limitations

The executable registry uses the provider alias mistral-large-latest; exact alias behavior can change as Mistral promotes GA versions.
Native document support differs from image support in this gateway adapter.

Cost

Consult Mistral's current pricing page for the exact API alias.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Depends on hosted endpoint, context, and generation length.

Privacy and data handling

Input path: Sent to the configured Mistral OpenAI-compatible endpoint.

Training and retention: Retention and training treatment depend on the Mistral account or endpoint operator and current terms.

Sensitive data: Use gateway redaction and verify the endpoint's enterprise data controls.

Privacy boundary: Compatibility metadata expresses intent only; it is not proof of provider retention behavior.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models
Mistral model lifecycle policy (official, reviewed 2026-09-13) · https://docs.mistral.ai/inference/model-lifecycle
Mistral model guide (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/models
Mistral pricing (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/pricing
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models

mistral-embedmistralcurrent familyexternal-educationexternal-education

embeddings · reviewed 2026-09-13 · next review 2026-10-13

Tasks

semantic search · retrieval · classification

Limitations

No embedding adapter exists in this gateway.
Embedding dimensions and index compatibility must be verified.

Cost

Consult Mistral's current inference pricing for embedding units.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Batching and endpoint load determine throughput; no Trexia benchmark is asserted.

Privacy and data handling

Input path: Sent to the configured Mistral OpenAI-compatible endpoint.

Training and retention: Retention and training treatment depend on the Mistral account or endpoint operator and current terms.

Sensitive data: Use gateway redaction and verify the endpoint's enterprise data controls.

Privacy boundary: Compatibility metadata expresses intent only; it is not proof of provider retention behavior.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models
Mistral model lifecycle policy (official, reviewed 2026-09-13) · https://docs.mistral.ai/inference/model-lifecycle
Mistral model guide (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/models
Mistral pricing (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/pricing
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models

Voxtral familymistralcurrent familyexternal-educationexternal-education

speech, text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

speech-to-text · audio understanding · meeting transcription

Limitations

No speech adapter exists in this gateway.
Audio language, diarization, and long-file limits require provider verification.

Cost

Consult Mistral's current audio pricing and model documentation.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Streaming and file transcription have different latency profiles.

Privacy and data handling

Input path: Sent to the configured Mistral OpenAI-compatible endpoint.

Training and retention: Retention and training treatment depend on the Mistral account or endpoint operator and current terms.

Sensitive data: Use gateway redaction and verify the endpoint's enterprise data controls.

Privacy boundary: Compatibility metadata expresses intent only; it is not proof of provider retention behavior.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models
Mistral model lifecycle policy (official, reviewed 2026-09-13) · https://docs.mistral.ai/inference/model-lifecycle
Mistral model guide (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/models
Mistral pricing (official, reviewed 2026-09-13) · https://docs.mistral.ai/getting-started/pricing
Mistral model catalog (official, reviewed 2026-09-13) · https://docs.mistral.ai/models

Qwen3-VL familyqwen3.xexternal-educationexternal-education

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

visual reasoning · OCR · agentic computer vision

Limitations

Not registered; registry contains Qwen 2.5 text and VL aliases.
Hosted aliases, snapshots, and regional availability differ.

Cost

Consult Alibaba Cloud Model Studio pricing for the exact region and alias.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Depends on visual token count, endpoint region, and reasoning settings.

Privacy and data handling

Input path: Sent to the configured Qwen-compatible endpoint.

Training and retention: Retention and training treatment depend on the Alibaba Cloud or self-hosted endpoint and account terms.

Sensitive data: Confirm geography, endpoint operator, and logging policy before sending confidential data.

Privacy boundary: Model availability and data-processing authorization are separate questions.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Alibaba Cloud Model Studio model catalog (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/models
Alibaba Cloud Model Studio pricing (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/model-pricing
Alibaba Cloud model lifecycle updates (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/newly-released-models
Alibaba Cloud model lifecycle updates (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/newly-released-models

Qwen3-VL-Embeddingqwen3.xexternal-educationexternal-education

embeddings, text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

multimodal retrieval · cross-modal search · reranking

Limitations

No Qwen embedding adapter exists in this gateway.
Vector index dimensions and modality alignment require a dedicated integration.

Cost

Consult Alibaba Cloud or the selected self-hosted endpoint pricing.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Batch size and image processing affect throughput.

Privacy and data handling

Input path: Sent to the configured Qwen-compatible endpoint.

Training and retention: Retention and training treatment depend on the Alibaba Cloud or self-hosted endpoint and account terms.

Sensitive data: Confirm geography, endpoint operator, and logging policy before sending confidential data.

Privacy boundary: Model availability and data-processing authorization are separate questions.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Alibaba Cloud Model Studio model catalog (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/models
Alibaba Cloud Model Studio pricing (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/model-pricing
Alibaba Cloud model lifecycle updates (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/newly-released-models
QwenLM repositories (official, reviewed 2026-09-13) · https://github.com/orgs/QwenLM/repositories

Qwen3-ASRqwen3.xexternal-educationexternal-education

speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

speech recognition · multilingual transcription · audio classification

Limitations

No speech adapter exists in this gateway.
Audio language and diarization support must be verified per endpoint.

Cost

Consult the selected Alibaba Cloud or self-hosted endpoint pricing.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Streaming/file mode and audio length determine latency.

Privacy and data handling

Input path: Sent to the configured Qwen-compatible endpoint.

Training and retention: Retention and training treatment depend on the Alibaba Cloud or self-hosted endpoint and account terms.

Sensitive data: Confirm geography, endpoint operator, and logging policy before sending confidential data.

Privacy boundary: Model availability and data-processing authorization are separate questions.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Alibaba Cloud Model Studio model catalog (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/models
Alibaba Cloud Model Studio pricing (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/model-pricing
Alibaba Cloud model lifecycle updates (official, reviewed 2026-09-13) · https://help.aliyun.com/en/model-studio/newly-released-models
QwenLM repositories (official, reviewed 2026-09-13) · https://github.com/orgs/QwenLM/repositories

meta/llama-4-maverick-instructnvidiaLlama 4external-educationexternal-education

text, vision · reviewed 2026-09-13 · next review 2026-10-13

Tasks

multimodal reasoning · visual extraction · agentic workflows

Limitations

NIM availability depends on the deployed container/catalog entry.
No matching model is registered in this repository catalog.

Cost

NIM self-hosted cost is infrastructure-dependent; hosted endpoint pricing is deployment-specific.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Depends on GPU class, batching, quantization, and image tokens.

Privacy and data handling

Input path: Sent to the NVIDIA NIM OpenAI-compatible endpoint configured by the operator.

Training and retention: The NIM deployment operator controls storage, logging, and training treatment; hosted NIM terms may differ from self-hosted NIM.

Sensitive data: Confirm the NIM endpoint's logging and isolation posture before sending PII or documents.

Privacy boundary: The compatibility payload can carry privacy hints, but enforcement belongs to the endpoint operator.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
NVIDIA NIM documentation (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/index.html
NVIDIA NIM LLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/large-language-models/1.4.0/models.html
NVIDIA NIM VLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/vision-language-models/latest/models.html
NVIDIA NIM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/vision-language-models/latest/models.html

NVIDIA NIM embedding familiesnvidiacurrent catalogexternal-educationexternal-education

embeddings · reviewed 2026-09-13 · next review 2026-10-13

Tasks

retrieval · reranking · semantic search

Limitations

The current gateway NIM adapter is chat-completion oriented and has no embedding route.
Container/model choice determines dimensions and license.

Cost

Self-hosted NIM cost is GPU/infrastructure-based; hosted pricing is deployment-specific.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Throughput depends on GPU, batching, and vector dimension.

Privacy and data handling

Input path: Sent to the NVIDIA NIM OpenAI-compatible endpoint configured by the operator.

Training and retention: The NIM deployment operator controls storage, logging, and training treatment; hosted NIM terms may differ from self-hosted NIM.

Sensitive data: Confirm the NIM endpoint's logging and isolation posture before sending PII or documents.

Privacy boundary: The compatibility payload can carry privacy hints, but enforcement belongs to the endpoint operator.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
NVIDIA NIM documentation (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/index.html
NVIDIA NIM LLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/large-language-models/1.4.0/models.html
NVIDIA NIM VLM model cards (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/vision-language-models/latest/models.html
NVIDIA NIM documentation (official, reviewed 2026-09-13) · https://docs.nvidia.com/nim/index.html

microsoft/Phi-4-multimodal-instructphiPhi-4 multimodalexternal-educationexternal-education

text, vision, speech · reviewed 2026-09-13 · next review 2026-10-13

Tasks

small-model multimodal reasoning · image understanding · audio understanding

Limitations

The executable registry routes image requests to microsoft/Phi-4-multimodal-instruct when the Phi adapter is selected.
Open-model weights do not provide a managed endpoint or retention guarantee.

Cost

Self-hosted cost is infrastructure-dependent; hosted endpoint pricing is provider-specific.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Small-model latency depends on hardware, quantization, and media preprocessing.

Privacy and data handling

Input path: Sent to the configured Phi-compatible endpoint.

Training and retention: The endpoint operator controls retention and training use; the Microsoft model card describes weights, not this deployment's inference policy.

Sensitive data: Use gateway rails and confirm operator controls for any sensitive input.

Privacy boundary: An open model does not itself provide a data-retention guarantee.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Microsoft Phi product page (official, reviewed 2026-09-13) · https://azure.microsoft.com/en-us/products/phi
Microsoft Foundry Phi-4 multimodal catalog (official, reviewed 2026-09-13) · https://ai.azure.com/catalog/models/Phi-4-multimodal-instruct
Microsoft Foundry retired models (official, reviewed 2026-09-13) · https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/retired-models
Microsoft Phi-4 model card (official, reviewed 2026-09-13) · https://huggingface.co/microsoft/phi-4
Microsoft Phi product page (official, reviewed 2026-09-13) · https://azure.microsoft.com/en-us/products/phi

tiiuae/Falcon3 familyfalcon3.xexternal-educationexternal-education

text · reviewed 2026-09-13 · next review 2026-10-13

Tasks

open-weight chat · local inference · fine-tuning research

Limitations

Not registered; current registry aliases Falcon 180B and 40B.
Weights/license, serving stack, and endpoint controls are deployment-specific.

Cost

Self-hosted cost is infrastructure-dependent; no hosted price is assumed.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Hardware, quantization, context length, and batching dominate latency.

Privacy and data handling

Input path: Sent to the configured Falcon-compatible endpoint.

Training and retention: The endpoint operator controls retention and any training use; the model card does not establish this deployment's data policy.

Sensitive data: Treat self-hosted or third-party Falcon endpoints as unverified until their logging and access controls are reviewed.

Privacy boundary: Open-source weights do not make an inference endpoint private by default.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
TII Falcon-40B model card (official, reviewed 2026-09-13) · https://huggingface.co/tiiuae/falcon-40b
TII Falcon-180B model card (official, reviewed 2026-09-13) · https://huggingface.co/tiiuae/falcon-180B
TII Falcon model cards (official, reviewed 2026-09-13) · https://huggingface.co/tiiuae

FLUX image familyBlack Forest Labscurrent familyexternal-educationexternal-education

image · reviewed 2026-09-13 · next review 2026-10-13

Tasks

image generation · listing creative · image editing

Limitations

No FLUX adapter is configured in this repository.
License and hosted/self-hosted data handling differ by checkpoint and provider.

Cost

Provider-specific or infrastructure-based; consult the selected official endpoint/checkpoint terms.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Generation time varies with resolution, steps, hardware, and queue.

Privacy and data handling

Input path: Sent to the customer-supplied OpenAI-compatible base URL.

Training and retention: The customer/operator controls retention, logs, and training use; this catalog cannot verify them.

Sensitive data: Require a documented endpoint data policy before allowing regulated or confidential data.

Privacy boundary: BYOK proves credential selection, not endpoint authorization, isolation, or retention.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Black Forest Labs website (official, reviewed 2026-09-13) · https://blackforestlabs.ai/

Stable Image familyStability AIcurrent familyexternal-educationexternal-education

image · reviewed 2026-09-13 · next review 2026-10-13

Tasks

image generation · image editing · creative ideation

Limitations

No Stability AI adapter is configured in this repository.
Model/license and endpoint data policies vary by deployment.

Cost

Provider-specific or infrastructure-based; consult the selected official endpoint pricing.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Generation time varies with resolution, steps, hardware, and queue.

Privacy and data handling

Input path: Sent to the customer-supplied OpenAI-compatible base URL.

Training and retention: The customer/operator controls retention, logs, and training use; this catalog cannot verify them.

Sensitive data: Require a documented endpoint data policy before allowing regulated or confidential data.

Privacy boundary: BYOK proves credential selection, not endpoint authorization, isolation, or retention.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
Stability AI developer platform (official, reviewed 2026-09-13) · https://platform.stability.ai/

BAAI/bge-m3BAAIM3external-educationexternal-education

embeddings · reviewed 2026-09-13 · next review 2026-10-13

Tasks

multilingual retrieval · semantic search · hybrid retrieval

Limitations

No BGE adapter is configured in this repository.
Self-hosted vectors require an embedding worker and compatible index dimensions.

Cost

Self-hosted cost is infrastructure-based; no managed API price is assumed.

official-pricing-reference · USD · Provider-specific; see the official pricing source

Input uncached $n/a/1M · cached $n/a/1M · output $n/a/1M

Batch input $n/a/1M · batch output $n/a/1M

Latency

unknown · Depends on CPU/GPU, batching, sequence length, and quantization.

Privacy and data handling

Input path: Sent to the customer-supplied OpenAI-compatible base URL.

Training and retention: The customer/operator controls retention, logs, and training use; this catalog cannot verify them.

Sensitive data: Require a documented endpoint data policy before allowing regulated or confidential data.

Privacy boundary: BYOK proves credential selection, not endpoint authorization, isolation, or retention.

Execution class

Education-only reference. Listing this model does not add a provider, credential, route, fallback, or model override.

Adapter registered: no · authorization/connectivity verified: no

Workflow: None asserted · source: None

Credential requirement: No Trexia execution path is asserted by this entry; provider authorization and connection are unverified.

Lifecycle

Status: external-education · registry: not-registered

Last reviewed 2026-09-13; next review 2026-10-13

Review trigger: Recheck the provider's official catalog, availability, pricing, and deprecation notices.

Included for literacy and comparison only; it is not a configured Trexia route.

Authoritative and repository sources

Trexia AI gateway model registry (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRegistry.ts
Trexia AI gateway model router (repository, reviewed 2026-09-13) · lib/ai-gateway/src/modelRouter.ts
Trexia AI gateway provider adapters (repository, reviewed 2026-09-13) · lib/ai-gateway/src/providers.ts
BAAI BGE-M3 model card (official, reviewed 2026-09-13) · https://huggingface.co/BAAI/bge-m3

Model and tool guidance is educational and static. Trexia does not automatically select, compare, purchase, invoke, or switch models from this curriculum. Provider documentation, pricing, capabilities, and lifecycle status change. Check the reviewed date and authoritative source before relying on a catalog entry.

Start Your Journey

Anonymous lesson checks and quiz grades stay in this browser only when you opt in. Choose a Trexia plan to save progress to an authenticated Academy profile and continue in Training Coach.

I consent to save this role's anonymous lesson and quiz activity in local storage for a one-time authenticated import. This is separate from account consent; read the Terms of Service.Create account / choose a plan Open Training Coach after sign-in

Account creation includes plan selection. After sign-in, open /training-coach?tab=academy&role=brokerages to confirm import.

Already have an account? Sign in to resume in your Training Coach.

View AI Model Directory

This link joins the existing journey; it does not change onboarding state or grant certification.