As announced in early February, both free and paid users of the ChatGPT platform will experience an updated version that shows in more detail the reasoning steps the o3-mini model takes to arrive at an answer. In particular, subscribers to the premium plan with the “advanced reasoning” configuration will see this thought process more clearly.
The move is aimed at increasing transparency and responding to competitive pressure from rivals, especially Chinese AI company DeepSeek.
“We are introducing an updated version of the reasoning chain for o3-mini to help users better understand how the model arrived at its results,” said an OpenAI representative. “With this change, users will be able to easily follow the reasoning steps, thereby gaining more confidence in the AI’s answers.”
Previously, OpenAI provided only brief summaries of the reasoning process of its o3-mini models and their predecessors, the o1 and o1-mini. However, these summaries were sometimes inaccurate and reduced the ability to judge the reliability of the results. To maintain a competitive advantage, OpenAI used to refrain from revealing the full details of its thinking process.
However, under pressure from AI models like DeepSeek R1, which has made its reasoning process public and is highly regarded by many AI experts, OpenAI has found a balance. Instead of revealing it completely, o3-mini can now “think freely” and reorganize its reasoning into more detailed summaries while still ensuring confidentiality.
In addition, OpenAI also added a post-processing step to increase the safety and clarity of the output. This process includes reviewing the raw argument chain, removing irrelevant content, and simplifying complex ideas. Notably, this processing step also supports translating the argument chain into languages other than English, making the experience more friendly and accessible for global users.
“We are working on exposing more information about the model’s thought process,” said Kevin Weil, OpenAI’s product manager, in a recent Reddit Q&A. “While full disclosure could hurt competition, we understand that users, especially advanced users, want this. We will find the right balance.”
With these changes, OpenAI hopes to improve the user experience while promoting trust in the accuracy and transparency of future AI models.
