OpenAI reveals “chain of thought” in ChatGPT’s o3-mini model to match DeepSeek’s open-source hype

According to a new announcement by OpenAI, ChatGPT’s o3-mini model now reveals a step-by-step chain of thought. This update comes as OpenAI is facing increasing pressure from rival companies like DeepSeek.

Earlier on Thursday, OpenAI announced that the paid and free users of ChatGPT will receive an “updated chain of thought” process in response to prompts. The chain of thought shows the model’s reasoning used to find answers. Paid subscribers using o3-mini-high will also receive the readout.

O3-mini is a reasoning model, and hence, it is expected to thoroughly fact-check its responses before displaying the result. This helps avoid fallacies or inaccuracies in responses. However, it comes with a trade-off: longer response times.

ChatGPT’s o3-mini model prompt (Source: OpenAI X account)

It looks like OpenAI is trying to catch up with DeepSeek’s increasing hype around its open-source nature. AI researchers love transparency and argue that, unlike OpenAI, DeepSeek does not gate-keep the AI tech, which is the right approach.

Apart from being open-source, DeepSeek’s R1 reasoning model also shows the complete thought process, making the model easier to study. At the same time, it also helps with the user-experience, and allows the users to easily figure out if the model is going off-track.

The OpenAI team is trying to make ChatGPT more transparent

The team has been planning this update for a while. Last week, on 1st February, the OpenAI team, including Sam Altman and Kevin Weil, hosted a Reddit AMA.

One reddit user asked, “Can we please see all of the thinking tokens?” In response, OpenAI VP Engineering responded, “We are looking into ways to show more of the thinking process. Stay tuned.” Kevin Weil also answered the question, and mentioned that the team is “working on showing a bunch more than we show today.”

“To improve clarity and safety, we’ve added an additional post-processing step where the model reviews the raw chain of thought, removing any unsafe content, and then simplifies any complex ideas.” – An OpenAI spokesperson told TechCrunch.

Hence, OpenAI is likely to adopt more transparency in the near future, even though ChatGPT still does not show the full reasoning steps for its o3-mini, o1, and o1-mini models. For now, it only displays a summary of the reasoning.

Source