OpenAI launches reasoning model o1 with visible thinking
On February 12, 2025, OpenAI announced the release of a new version of the o1 model, allowing users to see the model's 'thinking' process in real time. The feature, called visible chain-of-thought, is available in the ChatGPT interface for subscribers of Pro and Team plans.
How does visible chain-of-thought work?
The new feature displays the logical steps the model goes through before providing the final answer. Instead of just showing the result, the interface presents a reasoning sequence, allowing the user to follow how the model reached that conclusion. OpenAI states that the update aims to increase transparency and trust in the model's reasoning.
Availability and access
The functionality is available exclusively for subscribers of Pro and Team plans. Users on these plans can enable visible thinking display in ChatGPT settings. There is no forecast for expansion to other plans or the free version yet.
Implications for AI transparency
The measure is seen as an important step towards making AI models more understandable and auditable. Experts point out that the ability to inspect reasoning can help identify biases or logical errors, in addition to increasing user trust. However, there are also concerns about the depth of transparency, as language models do not reason exactly like humans.
OpenAI did not detail whether visible thinking will apply to all interactions or only specific questions, nor did it inform whether users can interrupt or correct the reasoning during the process.
Launch context
The release comes amid a growing debate about the need for artificial intelligence regulation and mechanisms to ensure safety and ethics in the use of generative models. OpenAI has invested in transparency features, such as model cards and impact reports, as a way to demonstrate commitment to responsibility.
For subscribers, the novelty is already available and can be tested directly in the ChatGPT interface. The company promises future updates based on user feedback.