Beginning on Friday, January 31, OpenAI is officially introducing its latest AI model, the o3-mini. The announcement was made in a blog post on their website, just over a month after its initial reveal during the “12 Days of OpenAI.”

As is customary with updated generative AI models, o3-mini presents enhancements over its predecessor, o1-mini. However, the improvements may not be as substantial as anticipated. According to OpenAI, both models deliver similar performances in fields such as mathematics, coding, and scientific reasoning. Notably, o3-mini has demonstrated a 24% increase in response speed during A/B testing, providing users with swifter answers. Testers have highlighted that o3-mini yields answers that are not only clearer but also exhibit improved reasoning capabilities. When challenged with medium levels of reasoning effort, o3-mini performs comparably to o1-mini in some assessments of reasoning and intelligence.

Both o1-mini and o3-mini are categorized as reasoning models, designed to process and analyze information before articulating responses. The o3-mini features three tiers of reasoning effort—low, medium, and high—tailored to different tasks. For example, in mathematical evaluations, the medium and high reasoning efforts of o3-mini surpass those of o1-mini, while its high-effort performance even exceeds the more robust variant of o1-mini. All three reasoning efforts outperform o1-mini specifically in advanced science queries, although o1 retains an advantage over them.

Effective immediately, o3-mini will take the place of the o1-mini model for all users. OpenAI has not explicitly stated the reasons for discontinuing access to o1-mini, but emphasizes that o3-mini comes with enhanced rate limits and reduced latency compared to its predecessor.

Initially, access to o3-mini will be exclusive to users of ChatGPT Plus, Team, and Pro plans, with Enterprise users slated to gain access within the week. Importantly, Plus and Team users will see their daily message limits increase from 50 on o1-mini to 150 on o3-mini.

Interestingly, free users will also have a chance to experience o3-mini but with certain restrictions. They can opt for the “Reasoning” mode in the message composer or regenerate a response. This marks the first time that free users are offered a reasoning model within ChatGPT, coinciding with Microsoft’s recent provision of o1’s reasoning capabilities to Copilot users at no cost.

For deeper insights into the o3-mini model, you can read more in our dedicated post. As the model begins its rollout today, its true performance will become clearer as users engage with it in real-world scenarios.