Add new vercel ai anthropic models

[Sameerlite]

Relevant issues

https://vercel.com/ai-gateway/models/claude-sonnet-4.5
Fixes #20743

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature

Changes

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 9, 2026 11:39am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR extends the model registry/pricing metadata with a set of new vercel_ai_gateway/anthropic/* entries (pricing, context windows, and capability flags), updating both the primary model_prices_and_context_window.json and the backup copy.

The change plugs into LiteLLM’s existing model-lookup mechanism: downstream routing and feature enablement (vision/tools/schema/caching) depend on these JSON entries being keyed by the exact model identifiers and having accurate capability flags.

Confidence Score: 3/5

• This PR is probably safe to merge after validating the new model identifiers and fixing the missing capability flag inconsistency.
• The change is limited to model metadata, but correctness hinges on exact model-key naming and accurate capability flags. One new entry (claude-sonnet-4.5) is inconsistent with the others, and the new key naming differs from existing Vercel Anthropic entries in the same file, which can break model lookup/routing if the IDs are not exact.
• model_prices_and_context_window.json

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds multiple Vercel AI Gateway Anthropic model entries with pricing/context. Notable inconsistency: claude-sonnet-4.5 lacks supports_response_schema, and new model key naming differs from existing claude-4-* Vercel Anthropic entries (verify IDs/aliases).
litellm/model_prices_and_context_window_backup.json	Mirrors the same new Vercel AI Gateway Anthropic model entries as the primary pricing file; inherits the same concerns about missing supports_response_schema for claude-sonnet-4.5 and potential model-id naming mismatches vs existing Vercel Anthropic keys.

Sequence Diagram

sequenceDiagram
  participant App as LiteLLM runtime
  participant Prices as model_prices_and_context_window.json
  participant Provider as Vercel AI Gateway

  App->>Prices: Lookup config by model key
  Prices-->>App: Return pricing + feature flags
  App->>App: Enable/disable features based on flags
  App->>Provider: Send request using selected model ID
  Provider-->>App: Response
  App-->>App: Apply schema/vision/tools handling

  Note over Prices,App: If model key is wrong/unrecognized,
  Note over Prices,App: lookup/routing may fail or use fallback.
  Note over Prices,App: If supports_response_schema missing,
  Note over Prices,App: schema features may be disabled for that model.

Loading

[greptile-apps]

_{2 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Missing response schema flag

vercel_ai_gateway/anthropic/claude-sonnet-4.5 omits "supports_response_schema": true even though the other newly-added vercel/anthropic models include it (and the PR description suggests these are Vercel AI Gateway Anthropic models with schema support). If LiteLLM uses this flag to gate JSON-schema / response-format features, this entry will behave inconsistently vs the rest. Add supports_response_schema here (or remove it from the others if it’s not actually supported).

[greptile-apps]

Possible bad model identifiers

These new keys use both claude-3-5-* / claude-3-7-* and claude-sonnet-4* naming, but the existing adjacent Vercel Anthropic entries use claude-4-opus / claude-4-sonnet (see vercel_ai_gateway/anthropic/claude-4-opus and .../claude-4-sonnet). If the runtime expects exact model IDs that match Vercel’s gateway (or has a normalization layer), these entries may be unreachable / silently misrouted. Please confirm these exact strings are valid model IDs in the gateway and align naming with the existing Vercel Anthropic conventions (or add aliases/normalization if required).

Also appears in: litellm/model_prices_and_context_window_backup.json:28606.