[Fix] Forward anthropic-beta headers to Bedrock and other anthropic providers

[JVenberg]

Title

[Fix] Forward anthropic-beta headers to Bedrock and other cross-providers

Relevant issues

Fixes:

• [Bug]: Anthropic Beta Headers Not Forwarded to Vertex AI Claude Models #15299
• [Bug]: anthropic-beta Headers Not Forwarded to Bedrock Provider #15622

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
• I have added a screenshot of my new test passing locally
• 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

Type

🐛 Bug Fix

Changes

Problem

The ProviderSpecificHeader mechanism in the LiteLLM proxy was only supporting single provider matching. When users sent requests with anthropic-beta headers (e.g., context-1m-2025-08-07 for extended context windows) through the proxy to AWS Bedrock, the headers were not forwarded because:

1. Headers were stored with custom_llm_provider="anthropic"
2. When Bedrock made the actual request, it looked for custom_llm_provider="bedrock"
3. The string mismatch caused headers to be dropped

This affected any cross-provider header scenarios where the same header needs to work across multiple providers (Anthropic, Bedrock, Vertex AI, etc.).

Solution

Enhanced the ProviderSpecificHeader mechanism to support multiple providers using a comma-separated format:

1. Updated add_provider_specific_headers_to_request() in litellm/proxy/litellm_pre_call_utils.py:

• Changed storage from single provider to comma-separated list
• Now stores: custom_llm_provider="anthropic,bedrock,bedrock_converse,vertex_ai"
• Maintains backward compatibility with single-provider format

2. Updated get_provider_specific_headers() in litellm/litellm_core_utils/get_provider_specific_headers.py:

• Modified retrieval logic to split by comma and check if current provider is in the list
• Simplified logic while maintaining existing behavior

3. Added comprehensive test coverage:

• 6 new unit tests in tests/test_litellm/litellm_core_utils/test_provider_specific_headers.py
• 1 new integration test in tests/proxy_unit_tests/test_proxy_utils.py
• Updated existing test to match new multi-provider format
• Tests cover multi-provider matching, backward compatibility, and edge cases

Test Results

All 8 tests passing (100% success rate):

• Multi-provider matching scenarios
• Backward compatibility with single providers
• Edge cases (spaces, None values, no matches)

Impact

• Users can now send anthropic-beta headers through the proxy and have them correctly forwarded to Bedrock
• Solves extended context window scenarios (1M+ tokens)
• Maintains backward compatibility - existing single-provider configurations continue to work
• Minimal code changes (2 core files modified)
• Low risk due to comprehensive test coverage

Files Modified

• litellm/proxy/litellm_pre_call_utils.py (lines 1187-1207)
• litellm/litellm_core_utils/get_provider_specific_headers.py (lines 6-29)
• tests/test_litellm/litellm_core_utils/test_provider_specific_headers.py (new tests added)
• tests/proxy_unit_tests/test_proxy_utils.py (tests updated and added)

[vercel]

@JVenberg is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.

[ishaan-jaff]

I think this is close, happy to merge and make sure it passes