Illustration: Sarah Grillo/Axios OpenAI on Friday released o3-mini, an AI model that focuses on improved reasoning capabilities at a more modest cost than its predecessor.
Why it matters: The announcement of the new model follows Chinese AI startup DeepSeek’s release of its R1 open-source model , which shook up the AI industry and heated up the U.S.-China AI arms race.
Driving the news: The o3-mini model, previewed last month , promises faster response times compared to o1, reduced computational requirements and the ability to handle simpler queries with ease, all with a lower cost. It’s the first time OpenAI has brought reasoning capabilities to users of its free ChatGPT service.
The o3-mini model is a smaller version of OpenAI’s o3 model, which has not yet been released for broad use.
o3-mini also offers the ability to choose between three levels of reasoning: low, medium, and high. ChatGPT Plus users will be able to send 150 queries a day to o3-mini.
The free tier of ChatGPT will also receive access to o3-mini, but it’s unclear how many queries non-paying users will be able to send each day. ChatGPT Enterprise and Edu customers will get access next week.
Zoom in: OpenAI also added a "think" option for GPT-4o and other models that shows up between the files and web search icons. When toggled on, it uses Chain-of-Thought (CoT) reasoning for responses, narrating the steps in its response to a query — similarly to the o1 model and DeepSeek’s R1. CoT reasoning prompts LLMs to generate intermediate reasoning steps before arriving at a final answer, similar to how our cognitive processes break down complex problems into smaller, manageable parts.
What they’re saying: "Our mission is to make sure the U.S. wins this race," OpenAI chief product officer Kevin Weil told CNBC on Thursday. "I want an AI that behaves according to democratic values," he said. "I don’t want an AI that behaves according to authoritarian values."
The big picture: OpenAI continues to raise more money this year and is said to be in talks with SoftBank for a new investment in […]
OpenAI releases o3-mini reasoning model following DeepSeek frenzy
