On Wednesday, OpenAI launched its latest reasoning models, o3 and o4-mini. As with its other o-series models, OpenAI’s o3 and o4-mini think for a longer period of time before responding in order to gather as much relevant data as possible.
OpenAI claims that these are the smartest and most capable AI models it has released to date. These are the first of OpenAI’s models to use and combine every feature in ChatGPT, including searching the web, analyzing uploaded files and other data, reasoning deeply about images, as well as generating images on the fly.
Depending on what you ask the chatbot to do, o3 and o4-mini will know when to utilize which of the tools in order to provide the most useful response to your request. OpenAI is one step closer to providing a fully agentic ChatGPT that can perform tasks for you.
According to OpenAI, o3 is better than older models at coding, math, science, visual perception, and more. It makes 20% fewer major errors than o1 did on challenging, real-world tasks, and o3 is a more reliable partner for generating and evaluating hypotheses.
Meanwhile, OpenAI refers to o4-mini as “a smaller model optimized for fast, cost-efficient reasoning.” o4-mini scored 99.5% on AIME 2025, outperforms o3-mini on non-STEM tasks and data science, and supports higher usage limits than o3.
As of today, o3, o4-mini, and o4-mini-high will officially replace o1, o3‑mini, and o3‑mini‑high in the model selector for ChatGPT Plus, Pro, and Team users. Starting next week, Enterprise and Edu users will gain access, and free users can try o4-mini by selecting ‘Think’ in the composer before submitting their request in ChatGPT.
Furthermore, OpenAI expects to launch o3-pro “in a few weeks with full tool support.”
