Skip to main content
Reasoning tokens count toward output usage. Use lower effort for faster, shorter responses. Use reasoning_effort in Chat Completions, reasoning.effort in Responses, and output_config.effort in Messages. none disables reasoning where supported. GLM 5.3 always reasons: disable-shaped controls select low effort and hide the trace. Hiding reasoning does not remove its token cost. Reasoning appears in message.reasoning_content for Chat, reasoning output items for Responses, and thinking blocks for Messages. The Playground displays reasoning separately from the answer.