OpenAI Launches o3 pro: Its Most Powerful Reasoning AI Model
OpenAI has officially released o3-pro, a high-performance AI model designed for advanced reasoning and complex tasks.
Now its available to ChatGPT Pro and Team users, o3-pro is also accessible via API with pricing geared toward professional and enterprise-grade applications.
What is o3-pro?
The o3-pro model is built on the same base as o3, OpenAI’s April 2025 reasoning model, but runs in a high-compute mode.
This means it uses more computational resources and allows extended thinking time, enabling it to tackle harder problems with deeper accuracy and insight.
OpenAI describes o3-pro as its most capable reasoning model to date, outperforming o3 across domains like: Science & mathematics,Education & tutoring, Programming & code generation, Data analysis & research,Content writing.
Performance Benchmarks
In the AIME 2024 test, o3-pro scored 93%; in GPQA Diamond, o3-pro achieved 84%, and in Codeforces, the o3-pro model got a ranking of 2,748. In all these benchmarks, o3-pro outperformed its underlying o3 model.
Availability & Pricing
Now Available to ChatGPT Pro and Team users and also via API access and Enterprise and Education plans will gain access next week.
API Pricing:
It costs $20 for input and $80 for output, per 1 million tokens.
You can process up to 200,000 tokens in a single context window, and the knowledge cutoff date is June 1, 2024.
Speed vs. Accuracy Trade-Off
While o3-pro excels at producing high-quality, logically sound answers, it does so at a slower pace than faster models like GPT‑4o or o4‑mini.
OpenAI notes that response times may take several seconds to minutes, especially for large or complex tasks.
Additionally, due to its size and power, Canvas and temporary chat features are currently disabled for o3-pro in ChatGPT.
News Gist
OpenAI has launched o3-pro, its most powerful AI for advanced reasoning. Available to ChatGPT Pro, Team, and API users, it excels in complex tasks but trades speed for deeper accuracy.
It outperforms previous models across multiple benchmarks.