fix(anthropic): set supports_web_search on current Claude models priced for web search

[devin-ai-integration]

Relevant issues

Fixes #33919

Linear ticket

Pre-Submission checklist

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

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Screenshots / Proof of Fix

Every current-generation Anthropic-direct entry already carries search_context_cost_per_query (the billing metadata for Anthropic's web_search_20250305 tool), yet omitted supports_web_search, so litellm.supports_web_search(...) returned False and Router web-search filtering excluded the deployments.

Before (base litellm_internal_staging), the issue repro printed False for all listed models except claude-3-7-sonnet-20250219

After (commit 68a3a56f70cee50d45cc5d7c0e6a423876c07674):

$ LITELLM_LOCAL_MODEL_COST_MAP=True python -c "
import litellm
for m in ['claude-sonnet-4-20250514','claude-sonnet-4-5','claude-sonnet-4-6','claude-sonnet-5','claude-opus-4-1','claude-opus-4-8','claude-fable-5','claude-3-7-sonnet-20250219']:
    print(m, litellm.supports_web_search(model=m))"
claude-sonnet-4-20250514 True
claude-sonnet-4-5 True
claude-sonnet-4-6 True
claude-sonnet-5 True
claude-opus-4-1 True
claude-opus-4-8 True
claude-fable-5 True
claude-3-7-sonnet-20250219 True

Live Anthropic web search confirming the flag is accurate (real API calls, commit 68a3a56f70cee50d45cc5d7c0e6a423876c07674):

$ python -c "
import litellm
for m in ['claude-sonnet-4-5','claude-opus-4-1']:
    r=litellm.completion(model='anthropic/'+m,
        messages=[{'role':'user','content':'What was a major world news headline today? Use web search.'}],
        tools=[{'type':'web_search_20250305','name':'web_search','max_uses':2}], max_tokens=1024)
    print(m, r.model_dump()['usage']['server_tool_use'])"
claude-sonnet-4-5 {'web_search_requests': 1, 'tool_search_requests': None}
claude-opus-4-1 {'web_search_requests': 1, 'tool_search_requests': None}

Type

🐛 Bug Fix

Changes

Added "supports_web_search": true to every Anthropic-direct (litellm_provider: anthropic) entry that already prices the web search tool via search_context_cost_per_query but was missing the capability flag; this covers claude-sonnet-4-20250514, claude-sonnet-4-5, claude-sonnet-4-6, claude-sonnet-5, claude-opus-4-1, claude-opus-4-1-20250805, claude-opus-4-20250514, claude-4-opus-20250514, claude-opus-4-5, claude-opus-4-5-20251101, claude-opus-4-6, claude-opus-4-6-20260205, claude-opus-4-7, claude-opus-4-7-20260416, claude-fable-5 and claude-opus-4-8. The same entries were updated in the bundled backup map so the two stay in sync

Added tests/test_litellm/test_anthropic_web_search_model_metadata.py. Beyond checking the specific models from the issue, it enforces the invariant that any Anthropic-direct entry priced for web search (search_context_cost_per_query present) must set supports_web_search, so future model additions can't silently reintroduce the drift; it also verifies the flag resolves through litellm.supports_web_search and that the backup map matches the canonical file

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

Link to Devin session: https://app.devin.ai/sessions/a3ffe74142e640118168500f7431c768

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This PR fixes a metadata gap where Anthropic-direct model entries priced for the built-in web-search tool (search_context_cost_per_query present) were missing "supports_web_search": true, causing litellm.supports_web_search(...) to return False and router web-search filtering to exclude those deployments.

• JSON fixes — "supports_web_search": true is added to 16 Anthropic-direct entries in both model_prices_and_context_window.json and its bundled backup, keeping the two files in sync.
• New test — test_anthropic_web_search_model_metadata.py verifies the specific issue models, enforces a forward-looking invariant (billing flag ⟹ capability flag, main file only), and spot-checks backup parity for the named models.
• Test isolation defect — test_supports_web_search_resolves_true directly assigns litellm.model_cost without using monkeypatch.setattr, so the overwritten module-level state is never restored and can affect subsequent tests in the session.

Confidence Score: 4/5

The JSON changes are correct and targeted; the test file introduces a module-level state mutation without rollback that can affect other tests in the same session.

The core fix — adding supports_web_search to the 16 affected model entries in both JSON files — is straightforward and correct. The new test's invariant check is a good guard for future regressions. The concern is in test_supports_web_search_resolves_true: litellm.model_cost is overwritten by direct assignment, which pytest's monkeypatch does not track and therefore does not restore after each parametrized run.

tests/test_litellm/test_anthropic_web_search_model_metadata.py — the litellm.model_cost assignment on line 45 should use monkeypatch.setattr for proper teardown.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds "supports_web_search": true to 16 Anthropic-direct entries that already carried search_context_cost_per_query; no unrelated models touched and no existing flags removed.
litellm/model_prices_and_context_window_backup.json	Mirror of main file changes — identical "supports_web_search": true additions for the same 16 models; backup is kept in sync.
tests/test_litellm/test_anthropic_web_search_model_metadata.py	New test file verifying the flag fix and a forward-looking invariant; litellm.model_cost is mutated without monkeypatch rollback (test isolation defect), and the invariant is not applied to the backup file.

_{Reviews (1): Last reviewed commit: "fix(anthropic): set supports_web_search ..." | Re-trigger Greptile}

[greptile-apps]

litellm.model_cost mutation not rolled back after test

litellm.model_cost is overwritten with a direct assignment, which monkeypatch does not track and therefore does not restore after the test finishes. Every test that runs later in the same pytest session will see the local-backup cost map instead of whatever was in litellm.model_cost before. Swapping to monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url="")) lets pytest restore the original value automatically.

[greptile-apps]

Invariant not enforced for the backup map

test_anthropic_entries_with_search_billing_advertise_web_search verifies that every anthropic-provider entry in the main file with search_context_cost_per_query also sets supports_web_search, but only checks model_prices_and_context_window.json. The backup file has an equivalent enforcement gap: test_backup_matches_main only spot-checks ISSUE_MODELS, so a future model added to the main file that accidentally omits supports_web_search in the backup would pass all tests.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_fix_anthropic_supports_web_search (68a3a56) with litellm_internal_staging (bd44c9e)}