Skip to content

Reasoning effort dropped, Claude Code

Quick fix

Wait for the next release containing the merged Responses proxy fix so reasoning.effort can be forwarded after model mapping.

Error output
Changing effort in the client has no effect.

This occurs when a Claude request is converted by CC Switch to the OpenAI Responses format. For `grok-4.7`, `supports_reasoning_effort()` does not include this model or its `grok-4.7-*` variants in its supported list. The conversion therefore omits `reasoning.effort`, causing xAI to use the default `high`.

A second variant exists in the same conversion path: in CC Switch `3.17.0`, an explicit `max` request from Claude Code is globally downgraded to `xhigh`. For the GPT-5.6 series, model-specific handling was proposed and implemented: preserve explicit `max`, while older models continue to fall back to `xhigh`.

  1. For `grok-4.7`, wait for the next release containing the merged #7369 fix; it forwards mapped `output_config.effort` as `reasoning.effort`.

  2. For an explicit `max` request in the GPT-5.6 series, use a version with the model-aware fix described in #7531; it preserves `max`, while older models remain compatible by falling back to `xhigh`.

ToolClaude Code
Version3.17.0
PlatformsLinux

(No FAQ yet)

This is an unofficial community wiki with no affiliation to the cc-switch authors or the project itself. Its content is compiled from the project's public GitHub issues. This site distributes no software.