OpenAI presented a new model of artificial intelligence – o3-pro, which the company calls its most powerful today. This is an updated version of the o3 model, focused on logical thinking. Unlike traditional AI models, it solves tasks step by step, which increases its effectiveness in areas where accurate reasoning is important, such as mathematics, physics and programming.
The o3-pro model is available for ChatGPT Pro and Team users, where it replaces the previous o1-pro. Enterprise and Edu users will get access to the new model next week. Also, it is already available for developers via API. The cost of use is $20 per million incoming tokens and $80 per million outgoing tokens. One million input tokens is roughly equal to 750,000 words.
o3-pro significantly outperforms o3 in all areas tested, especially in science, education, business, programming and writing. Experts praised her ability to precisely follow instructions, give clear, complete and accurate answers. The model supports the use of tools such as web search, file analysis, Python, image manipulation, and personalized memory.
At the same time, o3-pro works slower than o1-pro and has some limitations: temporary chats in ChatGPT are currently unavailable, the model does not create images and does not support Canvas. Despite this, o3-pro achieved high results in tests. On the AIME 2024 math exam, it outperformed Google’s top model, the Gemini 2.5 Pro, and on the GPQA Diamond science test, it beat Anthropic’s Claude 4 Opus. OpenAI positions o3-pro as the new standard for tasks that require high precision and deep understanding.
