Chained reasoning is a core feature of GPT-OSS that allows developers to adjust the reasoning strength based on task complexity:
- low intensity: for simple tasks such as quick quizzes
- medium intensity: Balancing performance and responsiveness
- high strength: suitable for mathematical proofs, complex logical reasoning and other scenarios
The configuration method is set via a system messagereasoning_effortparameters, for example:SystemContent.new().with_reasoning_effort('high'). High-intensity reasoning outputs a detailed analysis process containing theanalysischannel content (which needs to be parsed with the Harmony format), and the final answer will be given via thefinalchannel returns.OpenAI recommends that intermediate reasoning processes be hidden in production environments to prevent harmful information leakage.
This answer comes from the articleGPT-OSS: OpenAI's Open Source Big Model for Efficient ReasoningThe
May not be reproduced without permission:AI productivity tools " How is the chain inference feature of GPT-OSS implemented? How should developers configure it?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil