Meet o3‑pro, OpenAI's smartest AI model yet
What's the story
OpenAI has launched o3-pro, its most advanced artificial intelligence (AI) model to date.
It is an improved version of o3, a reasoning model that the company launched earlier this year.
The new model replaces the previous o1-pro for ChatGPT Pro and Team users this week. Enterprise and Edu users will get access next week.
Developers can also use it via the API at $20 per million input tokens and $80 per million output tokens—roughly 750,000 words per million input tokens.
AI evolution
o3-pro is a reasoning model
The o3-pro model is part of OpenAI's new class of "reasoning models."
These advanced AI systems solve problems step-by-step, making them more reliable in technical fields like physics, coding, and advanced mathematics.
As per OpenAI, internal evaluations show that it outperforms its predecessor o3 across all tested categories, especially in science, business, education, and writing.
Model features
Memory and other features are still under development
The o3-pro model has web search, Python code execution, file analysis, reasoning about visual inputs, and personalized responses via memory. However, it can't generate images yet.
It also doesn't support OpenAI's Canvas workspace feature and lacks temporary chat support in ChatGPT due to a "technical issue" being resolved by the company.
Benchmark results
Outperforms competitors in several benchmarks
In terms of performance, the o3-pro model appears to be leading the pack.
It has beaten Google's Gemini 2.5 Pro in AIME 2024 (a math benchmark) and outperformed Claude 4 Opus from Anthropic in GPQA Diamond, a test designed to evaluate PhD-level science knowledge.
However, the model's responses typically take longer than o1-pro to complete, according to OpenAI.