fix(vertex_ai): enable context-1m-2025-08-07 beta header

[edwiniac]

Problem

When using extra_headers: {anthropic-beta: context-1m-2025-08-07} with vertex_ai/claude-sonnet-4-6, the header is filtered out and Claude's 1M context window feature is unavailable.

Users get: prompt is too long: 460500 tokens > 200000 maximum

Root Cause

In anthropic_beta_headers_config.json, the context-1m-2025-08-07 header is set to null for vertex_ai, causing it to be filtered out by filter_and_transform_beta_headers().

Other providers (anthropic, azure_ai, bedrock, databricks) correctly have this set to "context-1m-2025-08-07".

Fix

Set context-1m-2025-08-07 to "context-1m-2025-08-07" for vertex_ai provider.

Changes

• litellm/anthropic_beta_headers_config.json: Changed context-1m-2025-08-07 from null to "context-1m-2025-08-07" for vertex_ai

Fixes #21861

[CLAassistant]

All committers have signed the CLA.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 22, 2026 3:04am

[greptile-apps]

Greptile Summary

This PR fixes a configuration issue where the context-1m-2025-08-07 Anthropic beta header was set to null for the vertex_ai provider in anthropic_beta_headers_config.json, causing it to be filtered out by filter_and_transform_beta_headers(). This prevented users from using Claude's 1M context window feature through Vertex AI, resulting in a prompt is too long: 460500 tokens > 200000 maximum error.

• Config fix: Changed context-1m-2025-08-07 from null to "context-1m-2025-08-07" for the vertex_ai provider, matching the behavior of anthropic, azure_ai, bedrock, and bedrock_converse providers
• No new tests added: The existing parametrized test suite in test_anthropic_beta_headers_filtering.py already validates that non-null headers are forwarded correctly for all providers including vertex_ai, so this change is covered by existing tests
• Minimal risk: Single-value change in a JSON configuration file with no code logic modifications

Confidence Score: 5/5

• This PR is safe to merge — it's a single JSON config value change that aligns vertex_ai with all other providers.
• The change is a one-line JSON config fix changing a null to a string value. It aligns the vertex_ai provider with all other providers (anthropic, azure_ai, bedrock, bedrock_converse, databricks) that already support this header. The existing test suite covers this behavior through parametrized tests. There are no code logic changes, no new dependencies, and no risk of regression.
• No files require special attention.

Important Files Changed

Filename	Overview
litellm/anthropic_beta_headers_config.json	Changes context-1m-2025-08-07 from null to "context-1m-2025-08-07" for the vertex_ai provider, enabling the 1M context window beta header to be forwarded instead of being filtered out. This aligns vertex_ai with anthropic, azure_ai, bedrock, and bedrock_converse providers which already support this header.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["User sends request with\nextra_headers: anthropic-beta: context-1m-2025-08-07"] --> B["filter_and_transform_beta_headers()"]
    B --> C{"Lookup header in\nvertex_ai config"}
    C -->|"Before fix: null"| D["Header filtered out ❌\n200k token limit enforced"]
    C -->|"After fix: context-1m-2025-08-07"| E["Header forwarded ✅\n1M context window enabled"]
    D --> F["Error: prompt too long\n460500 tokens > 200000 max"]
    E --> G["Request succeeds with\n1M token context window"]

Loading

_{Last reviewed commit: e2e0b40}

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}