2 min read
OpenAI has introduced o3-pro, a powerful upgrade to its existing o3 AI model, marking a significant step forward in the company’s ongoing pursuit of smarter, more reliable artificial intelligence.
Unlike traditional AI models, o3-pro builds on OpenAI’s reasoning-first architecture. Designed to work through problems step by step, this model shines in complex domains such as math, physics, programming, and more. The goal is improved consistency and accuracy, particularly in areas that require logical thinking and multi-step reasoning.
Exclusive Rollout and Pricing
The new model is immediately available for ChatGPT Pro and Team users, replacing the earlier o1-pro. Enterprise and Edu users will receive access in the following week. Developers can also start using o3-pro via API starting today.
API pricing is set at $20 per million input tokens and $80 per million output tokens. To put that in perspective, a million input tokens roughly translates to 750,000 words, a little more than the length of War and Peace.
OpenAI o3-pro is rolling out now to all Pro users in ChatGPT and in the API.
— OpenAI (@OpenAI) June 10, 2025
OpenAI boasts that in internal evaluations, human reviewers “consistently prefer o3-pro over o3 in every tested category“, especially when it comes to science, education, programming, business, and writing help. The model scores higher for clarity, comprehensiveness, instruction-following, and accuracy.
In terms of functionality, o3-pro supports an impressive suite of tools. It can:
- Search the web
- Analyze files
- Interpret visual inputs
- Run Python code
- Leverage memory for personalized responses
However, it’s not without limitations. OpenAI acknowledges that responses may take longer than with o1-pro. Additionally:
- Temporary chats are currently disabled due to a technical issue.
- Image generation is not supported.
- Canvas, OpenAI’s collaborative workspace feature, is not compatible with o3-pro at this time.
In terms of benchmark performance, o3-pro is a standout. On the AIME 2024 math exam benchmark, it outperforms Google’s Gemini 2.5 Pro. It also surpasses Claude 4 Opus by Anthropic on GPQA Diamond, a rigorous test for PhD-level science proficiency.
With o3-pro, OpenAI is doubling down on reasoning and reliability, raising the bar for what modern AI can achieve across a range of technical and academic tasks.
