OpenAI has made changes to how its o3-mini AI model presents its reasoning process. In response to competition from rivals like DeepSeek, the company is now revealing more details on how the model reaches its conclusions.
OpenAI announced that free and paid ChatGPT users will see an improved “chain of thought,” offering a clearer breakdown of the AI’s decision-making steps. Those using o3-mini in the “high reasoning” mode will also experience this enhancement. OpenAI believes this update will provide users with more clarity and confidence in the AI’s responses.
Reasoning models like o3-mini verify their own outputs before responding, reducing errors but increasing response times. Some AI researchers prefer models that show their full reasoning process, as seen with DeepSeek’s R1 model.
However, OpenAI has previously limited the visibility of these steps, offering only brief summaries that were sometimes inaccurate. Now, the company has struck a balance—allowing o3-mini to generate detailed reasoning summaries without exposing all its internal processes.
To ensure clarity and safety, OpenAI has added a post-processing step where the model reviews its reasoning, removes unsafe content, and simplifies complex ideas. This also enables non-English users to view the reasoning in their native language, improving accessibility.
Kevin Weil, OpenAI’s chief product officer, recently confirmed in a Reddit AMA that the company was working on expanding the visibility of its AI’s thought process. While OpenAI remains cautious about revealing too much due to competitive concerns, this update marks a significant step toward greater transparency and user understanding.
