docs: add thinking.summary field to /v1/messages and reasoning docs

[Chesars]

Relevant issues

Related to #21441

Pre-Submission checklist

• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

📖 Documentation

Changes

Adds documentation for the thinking.summary field:

1. anthropic_unified/index.md — Added summary as an optional sub-field of the thinking object in the Request Format section, with all 4 possible values ("auto", "concise", "detailed", "disabled") and a note about preservation when routing to non-Anthropic providers.

2. reasoning_content.md — Added a "Summary Preservation via /v1/messages Adapter" section documenting that thinking.summary is preserved when routing non-Claude models through the Anthropic messages adapter.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Error		Mar 5, 2026 1:05am

[greptile-apps]

Greptile Summary

Documents the thinking.summary field for the Anthropic messages adapter, following the code fix in PR #21441 that preserves user-specified summary values when routing to non-Anthropic providers.

• Added summary as an optional sub-field of thinking with supported values ("auto", "concise", "detailed", "disabled") to the unified Anthropic adapter docs
• Added a new "Summary Preservation" section with a code example showing how the summary value is forwarded when routing through non-Claude models
• Minor style issue: the code example uses an unusual max_tokens value that is likely a typo for 8192

Confidence Score: 4/5

• This is a low-risk documentation-only PR that is safe to merge.
• Documentation-only change with accurate content that aligns with the code fix in PR fix(anthropic): preserve thinking.summary when routing to OpenAI Responses API #21441. One minor typo in a code example (8096 vs 8192) is the only issue found.
• No files require special attention — both are straightforward documentation updates.

Important Files Changed

Filename	Overview
docs/my-website/docs/anthropic_unified/index.md	Adds summary as an optional sub-field of the thinking object with valid enum values and a note about provider routing. Clean documentation addition.
docs/my-website/docs/reasoning_content.md	Adds section documenting thinking.summary preservation via the Anthropic messages adapter with a code example. Minor typo in the example max_tokens value.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Client sends request with\nthinking.summary='concise'"] --> B["litellm.anthropic.messages.acreate()"]
    B --> C{"Is provider Anthropic?"}
    C -->|Yes| D["Pass thinking object\ndirectly to Claude"]
    C -->|No| E{"Is provider OpenAI?"}
    E -->|Yes| F["Responses API Adapter\ntranslate_thinking_to_reasoning()"]
    E -->|No| G["Chat Completions Adapter\ntranslate_anthropic_thinking_to_reasoning_effort()"]
    F --> H["reasoning.summary = 'concise'\n(preserved from thinking.summary)"]
    G --> I["reasoning_effort with\nsummary preserved"]
    H --> J["OpenAI Responses API"]
    I --> K["Chat Completions API"]

Loading

_{Last reviewed commit: 57c0b46}