fix(test): unbreak the integration-cost and proxy_e2e_anthropic_messages CircleCI jobs on main

[devin-ai-integration]

TLDR

Problem this solves:

• integration-cost has been red on main since 1b305cd landed
• proxy_e2e_anthropic_messages_tests has been red on main since c19a199 landed
• Both are stale test data, not production bugs

How it solves it:

• Repin the Fireworks fallback cache-read case to the 50% default the code now applies
• Run the all-beta-headers Bedrock cases on Claude Fable 5.1, the newest model, which accepts every mapped beta

User Flow

Before: a maintainer looks at any main pipeline in CircleCI and both jobs are red for reasons unrelated to their change

1. They open the latest main pipeline at https://app.circleci.com/pipelines/gh/BerriAI/litellm?branch=main
2. integration-cost fails with x-litellm-response-cost 0.0012456 != expected 0.0021672 on the deepseek-v4p1-flash-fallback_cache_read_at_input_rate case
3. proxy_e2e_anthropic_messages_tests fails on both test_bedrock_invoke_messages_with_all_beta_headers cases with Bedrock 400 invalid beta flag
4. They have to scroll past both to find out whether their own change broke anything

After: the same two jobs are green on main, so a red job means something real

1. They open the latest main pipeline at https://app.circleci.com/pipelines/gh/BerriAI/litellm?branch=main
2. integration-cost passes: the renamed fallback_cache_read_at_half_input_rate case expects 0.0012456
3. proxy_e2e_anthropic_messages_tests passes: the two Bedrock cases hit Fable 5.1 with the full mapped beta list and get 200
4. Any red job is now attributable to the PR under test

Relevant issues

Root causes:

• Cost: fix(cost_calc): default fireworks cached input to the documented 50% discount when the map has no cache-read rate #41917 (1b305cd) made Fireworks default cached input to 50% of the input rate when the map has no cache_read_input_token_cost. The case added in test(integration): literal request/response cost tracking suite on a test-owned cost map #41328 still pinned cached tokens at the full input rate. The actual value in CI, 0.0012456, is exactly 640 * 1.5e-7 + 12288 * 7.5e-8 + 380 * 6e-7. test(integration): endpoint, breakdown component and failure support in the cost harness #41999 carries the same repin inside a much larger change; this PR lands just the repin so main goes green now, and test(integration): endpoint, breakdown component and failure support in the cost harness #41999 can drop that hunk on rebase
• Bedrock: fix(anthropic): register thinking-binding-controls-2026-08-01 in beta headers config #41203 (c19a199) mapped thinking-binding-controls-2026-08-01 for bedrock and bedrock_converse. That mapping is correct, Fable 5.1 needs it. Verified live against Bedrock us-east-1 today: the flag is rejected with invalid beta flag on Opus 4.5, Sonnet 4.5 and Sonnet 4.6, and accepted on Opus 4.7 and Fable 5.1, on both the invoke and converse paths. The test asserts "every mapped beta works on the model under test", so it has to run on a model where that holds, and the repo rule is to test on the latest model in the family anyway

Affected release

Linear ticket

Resolves LIT-8149

Pre-Submission checklist

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

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• 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)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Setup: python litellm/proxy/proxy_cli.py --config tests/proxy_e2e_anthropic_messages_tests/test_config.yaml --port 4000 with LITELLM_MASTER_KEY=sk-1234 and the CI Bedrock credentials for us-east-1. The cost case only runs inside the CircleCI integration harness (scripted upstream plus Postgres and Redis), so its Before is the CircleCI log and its After is the CI run on this PR

Before (f49fd22)

Bedrock invoke, every mapped bedrock beta

1. curl -s http://localhost:4000/v1/messages -H 'Authorization: Bearer sk-1234' -H 'Content-Type: application/json' -H 'anthropic-version: 2023-06-01' -H 'anthropic-beta: advanced-tool-use-2025-11-20,compact-2026-01-12,computer-use-2025-01-24,computer-use-2025-11-24,context-1m-2025-08-07,context-management-2025-06-27,effort-2025-11-24,fine-grained-tool-streaming-2025-05-14,thinking-binding-controls-2026-08-01,tool-search-tool-2025-10-19' -d '{"model":"bedrock-claude-opus-4.5","max_tokens":10,"messages":[{"role":"user","content":"Say hello"}]}'
2. 400 {"type":"error","error":{"type":"invalid_request_error","message":"litellm.BadRequestError: BedrockException - {\"message\":\"invalid beta flag\"}. Received Model Group=bedrock-claude-opus-4.5 ..."}}
3. Same request with only anthropic-beta: thinking-binding-controls-2026-08-01 also returns 400; each of the other nine flags alone returns 200

Bedrock converse, every mapped bedrock_converse beta

1. Same curl with anthropic-beta: computer-use-2025-01-24,computer-use-2025-11-24,context-1m-2025-08-07,effort-2025-11-24,fine-grained-tool-streaming-2025-05-14,structured-outputs-2025-11-13,thinking-binding-controls-2026-08-01 and "model":"bedrock-converse-claude-sonnet-4.5"
2. 400 ... BedrockException - {\"message\":\"The model returned the following errors: invalid beta flag\"}

Fireworks fallback cache-read cost (CircleCI job 2191123 on main)

1. pytest tests/integration/cost_calculation/test_cost_tracking.py via .circleci/scripts/run_integration.sh cost
2. AssertionError: fireworks_ai-accounts-fireworks-models-deepseek-v4p1-flash-fallback_cache_read_at_input_rate: x-litellm-response-cost 0.0012456 != expected 0.0021672

After (7966f50)

Bedrock invoke, every mapped bedrock beta

1. curl -s http://localhost:4000/v1/messages -H 'Authorization: Bearer sk-1234' -H 'Content-Type: application/json' -H 'anthropic-version: 2023-06-01' -H 'anthropic-beta: advanced-tool-use-2025-11-20,compact-2026-01-12,computer-use-2025-01-24,computer-use-2025-11-24,context-1m-2025-08-07,context-management-2025-06-27,effort-2025-11-24,fine-grained-tool-streaming-2025-05-14,thinking-binding-controls-2026-08-01,tool-search-tool-2025-10-19' -d '{"model":"bedrock-claude-fable-5.1","max_tokens":10,"messages":[{"role":"user","content":"Say hello and nothing else"}]}'
2. 200 {"model":"bedrock-claude-fable-5.1","id":"msg_bdrk_rgklbehc2r5cuzdiphgzpxlhrp2di5zidgqqimaouki7yhki5qia","type":"message","role":"assistant","content":[{"type":"text","text":"Hello"}],"stop_reason":"end_turn","usage":{"input_tokens":17,"output_tokens":5, ...}}

Bedrock converse, every mapped bedrock_converse beta

1. curl -s http://localhost:4000/v1/messages -H 'Authorization: Bearer sk-1234' -H 'Content-Type: application/json' -H 'anthropic-version: 2023-06-01' -H 'anthropic-beta: computer-use-2025-01-24,computer-use-2025-11-24,context-1m-2025-08-07,effort-2025-11-24,fine-grained-tool-streaming-2025-05-14,structured-outputs-2025-11-13,thinking-binding-controls-2026-08-01' -d '{"model":"bedrock-converse-claude-fable-5.1","max_tokens":10,"messages":[{"role":"user","content":"Say hello and nothing else"}]}'
2. 200 {"id":"chatcmpl-4b275756-c40b-4161-b092-9cd53b129717","type":"message","role":"assistant","model":"bedrock-converse-claude-fable-5.1","usage":{"input_tokens":17,"output_tokens":5},"content":[{"type":"text","text":"Hello"}],"stop_reason":"end_turn"}
3. pytest tests/proxy_e2e_anthropic_messages_tests/test_all_beta_headers.py -k bedrock: both cases PASSED against the live proxy

Fireworks fallback cache-read cost

1. data_errors() on the edited fixture returns empty, so the renamed case and its contracts.json entry are consistent
2. The integration-cost job on this PR is the live run; see the checks tab

Type

✅ Test

Caveats (if any)

Low

• Whether Bedrock accepts a beta is per model; if a future mapped beta is rejected on Fable 5.1, move the cases to the newest model again
• Cost hunk overlaps test(integration): endpoint, breakdown component and failure support in the cost harness #41999, which will need a trivial rebase

Link to Devin session: https://app.devin.ai/sessions/fa3712a4b31d4d26a1877f606396ebc9
Open in Devin Desktop: https://app.devin.ai/desktop/session/fa3712a4b31d4d26a1877f606396ebc9?variant=devin

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• 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]

Confidence Score: 5/5

The PR appears safe to merge because the changed expectations match current production behavior and preserve the intended integration coverage

Summary

This PR repairs two stale integration-test expectations without changing production behavior

• Repins the Fireworks fallback cache-read fixture to the current 50% input-rate default and synchronizes its contract identifier
• Moves Bedrock all-beta-header coverage to Claude Fable 5.1 for both Invoke and Converse configurations
• Keeps the existing response, usage, provider-mapping, and cost-accounting assertions intact

_{Reviews (1) · Last reviewed commit: "fix(test): run the all-beta-headers bedr..."}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[joshua-berri]

lgtm