OpenAI has released o3-mini, its latest AI reasoning model that offers improved performance in STEM fields while being more cost-effective than its predecessors. The model, which was previewed in December, is now available through both ChatGPT and OpenAI’s API services.
The new model demonstrates significant improvements over o1-mini, with OpenAI reporting 24% faster response times and a 39% reduction in major errors on complex questions. External testers preferred o3-mini’s responses over o1-mini 56% of the time. The model particularly excels in mathematics, coding, and scientific reasoning, matching or exceeding the performance of the full o1 model in these areas when using medium or high reasoning effort settings.
In a notable departure from previous releases, OpenAI is making o3-mini available to free ChatGPT users for the first time. This access can be achieved by selecting the “Reason” button in the chat bar or by regenerating responses. Paid users receive enhanced benefits, with ChatGPT Plus and Team users getting 150 messages per day, while Pro subscribers have unlimited access to both o3-mini and o3-mini-high variants.
For developers, o3-mini introduces new features including function calling, structured outputs, and developer messages. The model is priced at $0.55 per million cached input tokens and $4.40 per million output tokens, making it 63% cheaper than o1-mini. Developers can choose between three reasoning effort levels – low, medium, and high – to balance accuracy and speed based on their specific needs.
The model comes with some limitations: it does not support vision capabilities, and users requiring image analysis must continue using o1. OpenAI has implemented safety measures through “deliberative alignment,” training the model to reason about human-written safety specifications before responding to prompts. The company claims o3-mini surpasses GPT-4o on safety and jailbreak evaluations.
This release comes amid increasing competition in the AI field, particularly from companies like DeepSeek, which recently launched its open-source R1 reasoning model. While o3-mini maintains OpenAI’s proprietary approach, it represents the company’s effort to make advanced AI capabilities more accessible while maintaining competitive pricing and performance standards.
Sources: OpenAI, TechCrunch, The Verge, VentureBeat, Simon Willison