OpenAI announced the general availability of o3 mini reasoning model

OpenAI originally introduced the new o3 family of reasoning models in December 2023 as part of its '12 Days of OpenAI' event. The o3 series contains two models: o3 and o3 tiny. OpenAI announced today that o3-mini models are now available in ChatGPT and via the API for developers.

As previously revealed, the o3-mini model would allow users to choose between three reasoning levels: high, medium, and low. The Low level will be the fastest but less accurate, whilst the High level will be the slowest but most accurate.

Until now, OpenAI's o1 series reasoning models were exclusively available to ChatGPT paid-tier subscribers. With o3-mini, OpenAI extends reasoning skills to free ChatGPT users as well. OpenAI says that o3-mini with medium can match the OpenAI o1 model's math, coding, and science performance. The essential point here is that the o3-mini can deliver it at a far lesser cost and in less time.

ChatGPT Plus, Team, and Pro users can access OpenAI o3-mini today, while Enterprise plan subscribers will have access next week. OpenAI will limit these users to 150 messages per day using o3-mini. ChatGPT free users can access o3-mini via the new 'Reason' option in the message composer or by regenerating a response. Additionally, for the first time, OpenAI has linked web search material with its reasoning models. As a result, o3-mini can provide current responses, as well as links to pertinent web sites.

For developers, o3-mini now has full support for function calls, Structured Outputs, streaming, and developer messages. Developers can also access it using the Chat Completions API, the Assistants API, and the Batch API. In addition, the o3-mini is substantially less expensive than the o1 series devices. OpenAI claims that o3-mini will cost 63% less than OpenAI o1-mini and 93% less than o1.

The release of the o3 small model is a significant step towards democratising access to advanced reasoning skills for both ordinary consumers and developers.