OpenAI has announced the launch of two new AI models, o3 and o4-mini, designed to think more deeply before forming answers. The company calls o3 its most powerful model yet, as it performs well in math, programming, logic, science, and visual understanding. The o4-mini, on the other hand, is aimed at developers looking for a balance between price, performance, and speed.
These models are capable of using ChatGPT tools, including web browsing, Python execution, image generation and analysis. They are also accompanied by a variant called o4-mini-high, which spends more time generating responses for improved accuracy. All of these models are available to Pro, Plus, and Team users, as well as through a developer API.
The new models are part of OpenAI’s strategy to compete with Google, Meta, xAI, Anthropic, and DeepSeek. Although OpenAI was the first to release a reasoning model, o1, competitors quickly caught up.
o3 showed the highest result in the SWE-bench test without additional support — 69.1%, while o4-mini reached 68.1%. For comparison, Claude 3.7 Sonnet scored 62.3%, and OpenAI's previous model — o3-mini — only 49.3%.
The models also have the ability to “think visually”: users can upload sketches, diagrams, or PDFs that the AI will analyze before responding. They can also work with blurry or low-quality images, and scale or rotate them as they reason.
As for the prices, they are: $10 per million input tokens and $40 per output for o3, and $1.10/$4.40 for o4-mini. In the future, there is also an expected release of o3-pro, available only to ChatGPT Pro users. By the way, these are probably the last reasoning models before the release of GPT-5.