PR #25667: fix: support reasoning_effort for Claude models on github_copilot provider

• Repository: BerriAI/litellm
• Author: anchenyi
• State: closed | merged: False
• Link: https://github.com/BerriAI/litellm/pull/25667

Description (problem / solution / changelog)

Relevant issues

Fixes https://github.com/BerriAI/litellm/issues/25666

Pre-Submission checklist

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

• I have Added testing in the tests/test_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

Screenshots / Proof of Fix

Before fix — reasoning_effort not present in request to Copilot API:

After fix — reasoning_effort: "medium" correctly forwarded to Copilot API, response contains reasoning output:

Type

🐛 Bug Fix

Changes

Problem

The github_copilot provider does not support reasoning_effort for Claude models. Two issues:

1. GithubCopilotConfig inherits OpenAIConfig.map_openai_params, which delegates to a hardcoded openAIGPTConfig singleton. That singleton uses its own get_supported_openai_params — not the subclass override — so reasoning_effort is filtered out.

2. When requests come from Anthropic-compatible clients (e.g., Claude Code via POST /v1/messages), the native thinking parameter is not converted to reasoning_effort. Copilot's OpenAI-compatible API does not understand thinking, so the intent is lost.

Fix

litellm/llms/github_copilot/chat/transformation.py:

• Override get_supported_openai_params to add thinking and reasoning_effort for all Claude models.
• Override map_openai_params to:
  • Convert Anthropic-native thinking → reasoning_effort (defaults to "medium" when thinking is enabled and no explicit reasoning_effort is set).
  • Strip thinking from the request (Copilot doesn't understand it).
  • Use self._map_openai_params() for Claude models instead of delegating to the singleton, so the correct get_supported_openai_params is called.
• Non-Claude models (GPT, O-series) are unaffected — they fall through to the parent implementation.

tests/test_litellm/llms/github_copilot/test_github_copilot_transformation.py:

• Added 11 new tests covering reasoning_effort pass-through, thinking → reasoning_effort conversion, thinking stripping, and non-Claude model behavior.

Changed files

• litellm/llms/github_copilot/chat/transformation.py (modified, +68/-10)
• tests/test_litellm/llms/github_copilot/test_github_copilot_transformation.py (modified, +146/-6)