OpenAI Unveils Next-Gen o3 Model Set for Early Next Year Release

OpenAI has previewed its next-generation frontier model, o3, promising improved performance and new features, with a mini version launching first.

OpenAI concluded its 12 Days of OpenAI livestream series by introducing the next-generation frontier model, o3, alongside a smaller version, o3-mini. The o3 model, named in a nod to the O2 cellular network and OpenAI’s naming tradition, is not yet available for public use. Initially, it will be accessible to researchers for safety testing purposes.

The o3-mini model is slated for release by the end of January, with o3 to follow shortly after. The o3 model boasts significant performance enhancements over its predecessor, o1, demonstrated by its 96.7% accuracy on the American Invitational Mathematics Examination, compared to o1’s 83.3%.

OpenAI has also introduced more challenging benchmarks, such as ARC-AGI, to test the model’s ability to intuit and learn spontaneously. The o3 model achieved a 75.7% score on ARC-AGI under low-compute settings and 87.5% with additional processing power, surpassing the human performance threshold of 85%.

The o3-mini model features the Adaptive Thinking Time API, offering three reasoning modes: Low, Medium, and High, allowing users to adjust the software’s problem-solving time. This model delivers results comparable to the current o1 model but at a reduced compute cost.