OpenAI has released o3-mini, providing a more powerful model to face off against the dark horse DeepSeek in what is shaping up to be an epic power struggle. The reasoning model is designed to be much more competent in coding, math, and science. OpenAI also launched Deep Research, which is meant to handle complex, multi-step research and which will undoubtedly become a favorite of research-averse students. The new model is available in ChatGPT for the web, as well as via the API.
“OpenAI o3-mini is our first small reasoning model that supports highly requested developer features including function calling, Structured Outputs, and developer messages, making it production-ready out of the gate,” OpenAI said in a statement.
If you have ChatGPT Plus, Team, or Pro, you can use OpenAI o3-mini right away. OpenAI is also providing Plus and Team users with 150 o3-mini messages per day, which is a noteworthy upgrade from the 50 o1-mini messages per day. You can also use it if you are a free user by picking Reason in the message composer, according to OpenAI. When you regenerate a ChatGPT response, it will also make use of o3-mini.
OpenAI’s Deep Research feature in ChatGPT.
Credit: OpenAI
O3-mini has been in the works for some time. OpenAI previewed the model in late December, long before DeepSeek made its big splash. Still, o3-mini represents a necessary response to DeepSeek, which appears to give surprisingly-good bang for the buck when it comes to AI hardware requirements. DeepSeek reportedly runs the model on cheaper Nvidia H800 chips designed for the Chinese market, thanks to restrictions by the US and other countries.
The new OpenAI release also includes Deep Research, which is an o3-powered tool that is expected to handle complicated research tasks. For example OpenAI’s preview of the tool uses the prompt: “Compile a research report on how the retail industry has changed in the past 3 years. Use bullets and tables where necessary for clarity.”
You can find Deep Research in ChatGPT’s message composer. OpenAI warns that the tool may need some time – even up to half an hour – to prepare research for you. The company expects to have Deep Research embedding images and other visuals in its reports in the coming weeks.
As crucial as performance is to the AI race, the businesses behind these models may also matter. OpenAI is owned by a combination of US groups, including the OpenAI non-profit and Microsoft. DeepSeek, on the other hand, was developed by a Chinese hedge fund and is designed to make use of older, less-powerful hardware that may not scale as well long-term. DeepSeek also sends loads of user data to China, raising privacy concerns.
The US government has committed to OpenAI, which launched ChatGPT Gov late last month, as The Verge points out. The US government won’t touch DeepSeek with a 10-foot pole, but it’s available to private citizens and businesses.