OpenAI showcases o3, o3-mini models with better reasoning capabilities and more

On the last day of its 12 Days of OpenAI livestream event, OpenAI revealed its latest advancements in artificial intelligence with the introduction of the o3 and o3-mini models. OpenAI CEO Sam Altman explained that the decision to name the models o3 and o3-mini was to avoid potential trademark conflicts with the British telecom provider O2. “In the grand tradition of OpenAI being really, truly bad at names, it’s called o3,” said Altman in a YouTube video.

According to Altman, these new models mark an important leap forward in AI reasoning capabilities, with the o3 model achieving an important milestone by cracking the ARC-AGI benchmark, a test designed to evaluate generalized intelligence.

The o3 model is a successor to the earlier o1 model and boasts improvements in reasoning and problem-solving abilities. According to OpenAI, the o3 model scored an impressive 87.5% on the ARC-AGI benchmark, surpassing the human average of 85%2.

In addition to the o3 model, OpenAI also introduced the o3-mini, a distilled version optimised for specific tasks such as coding. The o3-mini model offers faster performance and lower inference costs while maintaining high accuracy. It features three compute settings—low, medium, and high—allowing users to balance speed and performance based on their needs.