OpenAI has revealed its latest progress in artificial intelligence with the introduction of o3 and o3 mini, its next-generation reasoning models.
In a livestream event, Mark Chen, OpenAI’s Senior Vice President of Research, highlighted o3’s outstanding performance across various benchmarks. In comparison to the previous o1 model, o3 showed exceptional results in areas such as competitive math, attaining an impressive 96.7% accuracy, as well as PhD-level science, scoring 87.7%. Furthermore, o3 showcased its flexibility by obtaining a 76% score on the ARC-AGI benchmark, a challenging assessment created to test the ability to learn and implement new skills on unique, unpublished datasets.
This announcement marks the conclusion of OpenAI’s “12 Days of OpenAI” marathon, a sequence of daily launches that spotlight the company’s most recent innovations. Over the last 12 business days, OpenAI has rolled out a variety of groundbreaking tools and features, such as the AI video generator *Sora*, the *Advanced Voice Mode* for visual capabilities, and numerous updates designed to improve ChatGPT’s functionality in both professional and everyday settings. These initiatives correspond with OpenAI’s broader mission of transforming ChatGPT into a versatile, comprehensive application.
The o3 mini model, a more budget-friendly version of o3, is crafted to provide a balance between performance and cost-effectiveness. It offers three effort levels and can adapt its reasoning time on the complexities of specific problems. OpenAI CEO Sam Altman referred to it as delivering “an incredible cost-to-performance gain.”
Though o3 and o3 mini signify noteworthy advancements in AI intelligence, they are not yet open for public usage. However, OpenAI is launching an early access program for safety testing, commencing today. Those interested can apply to participate in the testing program, with applications accepted on a rolling basis until January 10. For further details, please visit OpenAI’s [early access program page](https://openai.com/index/early-access-for-safety-testing/).
These advancements highlight OpenAI’s dedication to expanding the frontiers of AI technology while prioritizing safety and reliability through thorough testing.