fix: allow effort="max" for Claude Opus 4.6 adaptive thinking

[KristofferArlind]

Relevant issues

Fixes effort="max" not being accepted for Claude Opus 4.6 in output_config validation. This is allowed for Opus 4.6, but not 4.5, as described here: https://platform.claude.com/docs/en/build-with-claude/effort.

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
• 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
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🐛 Bug Fix

Changes

• Added max as a valid effort level in output_config for Claude Opus 4.6
• Added validation so that effort='max' is rejected for other models
• Added test_effort_validation_with_opus_46 and test_max_effort_rejected_for_opus_45 tests

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 13, 2026 10:41am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR adds support for effort="max" in output_config validation for Claude Opus 4.6, which is documented as a valid value only for that model. The implementation expands the allowed effort values from ["high", "medium", "low"] to include "max", and adds a model-specific guard using the pre-existing _is_claude_opus_4_6() helper so that effort="max" is rejected for all other models (Opus 4.5, Sonnet, Haiku, etc.).

• Added "max" to the valid effort values in output_config validation in AnthropicConfig.transform_request()
• Added model-specific check: effort='max' raises ValueError for non-Opus 4.6 models
• Added two mock-only tests covering both the happy path (Opus 4.6 accepts all four effort levels) and the error path (Opus 4.5 rejects effort='max')
• Minor note: the pre-existing _is_claude_opus_4_6 method has an incorrect docstring ("Check if the model is Claude Opus 4.5") that should be corrected

Confidence Score: 4/5

• This PR is safe to merge — it adds a narrowly scoped validation change with proper tests and no risk of regression.
• The change is minimal and well-targeted: it expands a validation list and adds a model-specific guard. The logic is correct and the tests cover both positive and negative cases. The only minor issue is a pre-existing incorrect docstring on the helper method being relied upon. Provider-specific code stays within the llms/ directory per project conventions.
• No files require special attention. The pre-existing docstring on _is_claude_opus_4_6 at line 175 of litellm/llms/anthropic/chat/transformation.py should be corrected but is not a blocking issue.

Important Files Changed

Filename	Overview
litellm/llms/anthropic/chat/transformation.py	Adds "max" to the valid effort levels in output_config validation and adds a model-specific guard so only Claude Opus 4.6 can use effort='max'. Logic is correct. Minor note: pre-existing docstring on _is_claude_opus_4_6 incorrectly says "Opus 4.5".
tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py	Adds two well-structured mock-only tests: one verifying all four effort levels pass for Opus 4.6, and one verifying effort='max' is rejected for Opus 4.5. No real network calls.

Sequence Diagram

sequenceDiagram
    participant Client
    participant AnthropicConfig as AnthropicConfig.transform_request()
    participant Validator as Effort Validator
    participant ModelCheck as _is_claude_opus_4_6()
    participant API as Anthropic API

    Client->>AnthropicConfig: transform_request(model, output_config={effort: "max"})
    AnthropicConfig->>Validator: Check effort value
    Validator->>Validator: effort in ["high","medium","low","max"]?
    alt Invalid effort value
        Validator-->>Client: ValueError("Invalid effort value")
    end
    Validator->>ModelCheck: effort == "max"?
    ModelCheck->>ModelCheck: "opus-4-6" in model?
    alt Model is NOT Opus 4.6
        ModelCheck-->>Client: ValueError("effort='max' only supported by Opus 4.6")
    end
    ModelCheck-->>AnthropicConfig: Validation passed
    AnthropicConfig->>API: Send request with output_config

Loading

_{Last reviewed commit: 5aa3bfe}

[greptile-apps]

_{2 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (1)

litellm/llms/anthropic/chat/transformation.py
Incorrect docstring on pre-existing method
The docstring says "Check if the model is Claude Opus 4.5" but the method checks for Opus 4.6 ("opus-4-6" / "opus_4_6"). This was a pre-existing copy-paste error, but since this PR relies on this method for the new effort='max' validation, it's worth fixing to avoid confusion.

        """Check if the model is Claude Opus 4.6."""