the OpenAI o3 mini, which was previewed in December of 2024. It is part of its reasoning series “most cost-efficient model” . The o3 mini is optimized for STEM reasoning and excels in science, math and coding.
OpenAI claims that o3 mini is a specialized option for tech domains requiring speed and precision. With medium reasoning efforts, o3 mini matches the performance of o1 in science, math and coding, while delivering faster response times. The company said that in A/B tests, o3 mini was 24% faster than the o1 mini with an average response of 7.7s as opposed to 10.16s for o1mini.
OpenAI’s o3 mini is its first small reasoning model that supports highly requested developer features including function calling and developer messages. It also supports Structured Outputs. It will also support streaming and developers will be able to choose between three reasoning efforts: low, medium, or high. o3 Mini will also use search to find the most recent answers and provide links to relevant sources online.
ChatGPT Plus Team and Pro subscribers can already use the o3 mini. It can be chosen from the model selector, where it replaces the o1 mini. All paid users have the option of selecting “o3-mini-high” from the model picker, if they’re willing to sacrifice response time for greater intelligence. ChatGPT Pro users have unlimited access to o3 mini and o3 mini-high. Plus and Team users get a higher rate cap – from 50 messages/day for o1-mini, to 150 messages/day for o3mini. ChatGPT free users without a subscription can test o3 mini by selecting “Reason” from the message composer, or by regenerating an answer. This is OpenAI’s very first reasoning model available to ChatGPT free users. It is also available on Microsoft’s Azure OpenAI Service.
Source.
