feat(proxy): serve Anthropic-native /v1/models for Claude Code gateway discovery

[Ar-maan05]

Relevant issues

Fixes #27180

Pre-Submission checklist

• I have added meaningful tests
• My PR passes all unit tests on the affected directory
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Type

🆕 New Feature

Changes

Claude Code 2.1.126+ added gateway model discovery: when ANTHROPIC_BASE_URL points at a gateway, it queries {base_url}/v1/models at startup and populates the /model picker with the discovered models. That discovery only parses the Anthropic-native Models API shape, so against litellm (which returns OpenAI's {id, object, created, owned_by} list) Claude Code finds nothing and the picker stays empty, even though /v1/messages already works.

This serves the Anthropic-native shape from the same /v1/models route via content negotiation on the anthropic-version header. Claude Code already sends that header for /v1/messages, so when it is present the endpoint returns the Anthropic Models envelope (type / display_name / created_at per entry, plus top-level has_more / first_id / last_id); otherwise the response is byte-for-byte the existing OpenAI shape, so aider and other OpenAI-compatible clients are unaffected. A separate endpoint was not used because Claude Code discovers at the gateway root's /v1/models, and a global config flag would break the OpenAI clients that share the route.

The full model list is returned and Claude Code applies its own claude/anthropic id-prefix filter client-side, so no server-side filtering is imposed (a model aliased to claude-* that points at any backend still shows up, which is the point for gateway users). display_name falls back to the model id, the stable label a gateway can offer for arbitrary upstream models, and created_at is the ISO 8601 form of the same timestamp the OpenAI shape already returns. Hidden/unhealthy models are filtered before formatting in both the normal and scope=expand branches, exactly as for the OpenAI shape.

Tests

In tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py, two unit tests pin create_anthropic_model_list_response: the full envelope (per-entry type/display_name/created_at with a Z-suffixed ISO timestamp, top-level has_more/first_id/last_id, no object) and the empty-list case (first_id/last_id null). In tests/test_litellm/proxy/proxy_server/test_routes_models.py, a route test drives GET /v1/models (and /models) with the anthropic-version header and asserts the negotiated Anthropic shape, while the existing happy-path test pins that the default response stays OpenAI. Mutation-checked by hand: forcing the negotiation off, emitting object instead of type, and dropping the Z normalization each fail a test. The affected suites are green.

Screenshots / Proof of Fix

Live proxy on localhost:4000 with two claude-* deployments and one gpt-4o.

python litellm/proxy/proxy_cli.py --config litellm/proxy/dev_config_27180.yaml --port 4000

Default request (no header): OpenAI format, unchanged.

curl -s http://127.0.0.1:4000/v1/models -H "x-api-key: sk-1234" | python3 -m json.tool

{
    "data": [
        {"id": "claude-opus-4-6", "object": "model", "created": 1677610602, "owned_by": "openai"},
        {"id": "claude-haiku-4-5", "object": "model", "created": 1677610602, "owned_by": "openai"},
        {"id": "gpt-4o", "object": "model", "created": 1677610602, "owned_by": "openai"}
    ],
    "object": "list"
}

With the anthropic-version header (what Claude Code sends): Anthropic-native format.

curl -s http://127.0.0.1:4000/v1/models -H "x-api-key: sk-1234" -H "anthropic-version: 2023-06-01" | python3 -m json.tool

{
    "data": [
        {
            "type": "model",
            "id": "claude-opus-4-6",
            "display_name": "claude-opus-4-6",
            "created_at": "2023-02-28T18:56:42Z"
        },
        {
            "type": "model",
            "id": "claude-haiku-4-5",
            "display_name": "claude-haiku-4-5",
            "created_at": "2023-02-28T18:56:42Z"
        },
        {
            "type": "model",
            "id": "gpt-4o",
            "display_name": "gpt-4o",
            "created_at": "2023-02-28T18:56:42Z"
        }
    ],
    "has_more": false,
    "first_id": "claude-opus-4-6",
    "last_id": "gpt-4o"
}

The shape matches the Anthropic Models API. All models are returned and Claude Code keeps the claude/anthropic-prefixed ones for the picker; clients that do not send anthropic-version get the unchanged OpenAI list.

[greptile-apps]

Greptile Summary

This PR adds Anthropic-native model list content negotiation to the existing /v1/models endpoint: when the anthropic-version request header is present (as Claude Code sends for all Anthropic API calls), the endpoint returns the Anthropic Models API envelope (type/display_name/created_at per entry, plus has_more/first_id/last_id); all other callers continue to receive the unchanged OpenAI-compatible shape.

• create_anthropic_model_list_response is correctly placed in litellm/llms/anthropic/common_utils.py, satisfying the rule against provider-specific code outside of llms/; it reuses the existing DEFAULT_MODEL_CREATED_AT_TIME constant and produces a Z-suffixed ISO 8601 timestamp consistent with the Anthropic Models API.
• model_list in proxy_server.py gains a request: Request = None injection and a single wants_anthropic_format flag checked in both the scope=expand and normal branches — the existing OpenAI response path is untouched when the header is absent, so no backwards-incompatible change is introduced.
• Tests cover the full envelope shape, the empty-list case, and both affected HTTP routes, all fully mocked with no real network calls.

Confidence Score: 5/5

Safe to merge — the change is additive and strictly backward-compatible; callers without the anthropic-version header receive a byte-for-byte identical response to today.

The implementation is a clean, additive content-negotiation layer. Provider-specific formatting logic is correctly placed in the llms/anthropic/ module rather than shared proxy utilities. The request: Request = None default guards all non-HTTP call sites. Both affected branches in model_list handle the new flag consistently, tests are fully mocked, and the existing OpenAI-format path is untouched.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/anthropic/common_utils.py	Adds create_anthropic_model_list_response helper that builds the Anthropic-native /v1/models envelope; uses the existing DEFAULT_MODEL_CREATED_AT_TIME constant and ISO 8601 Z-suffix formatting. No fastapi imports, correctly placed in the provider-specific module.
litellm/proxy/proxy_server.py	Injects request: Request = None into model_list and performs content negotiation on the anthropic-version header in both the scope=expand and normal branches. Existing OpenAI-format response is byte-for-byte unchanged when the header is absent.
tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py	Adds two unit tests for create_anthropic_model_list_response: full envelope shape (type/display_name/created_at with Z suffix, has_more/first_id/last_id, no object) and empty-list case. All mocked, no network calls.
tests/test_litellm/proxy/proxy_server/test_routes_models.py	Adds test_get_models_anthropic_format_when_header_present that drives both /v1/models and /models with the anthropic-version header and pins the Anthropic-native shape; existing happy-path test is untouched, confirming OpenAI format is unchanged without the header.

_{Reviews (3): Last reviewed commit: "fix(proxy): make model_list request para..." | Re-trigger Greptile}

[codecov]

Codecov Report

❌ Patch coverage is 91.66667% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/proxy/proxy_server.py	83.33%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Ar-maan05]

@greptile-apps

[Ar-maan05]

Hi, my PR is passing CI and has a 5/5 score from greptile. Please let me know if any changes are required.

[Sameerlite]

@greptileai

[Ar-maan05]

Hi @Sameerlite, please let me know if any changes are needed. Always a pleasure to contribute to this repo.