OpenAI launches o3-pro, its most capable AI model yet: Key details

OpenAI

OpenAI has officially launched o3-pro, its most advanced model to date, aiming to set a new benchmark in AI reasoning. The upgrade replaces o1-pro for ChatGPT Pro and Team users starting this week, with Enterprise and Edu users to follow. Developers can also access it immediately via the API, where it’s priced at $20 per million input tokens and $80 per million output tokens—roughly 750,000 words per million input tokens.

The o3-pro model is part of OpenAI’s new class of “reasoning models”—AI systems designed to solve problems step-by-step, making them more reliable in technical domains like physics, programming, and advanced mathematics. Internal evaluations suggest it outperforms its predecessor o3 across all tested categories, particularly in science, business, education, and writing support. Reviewers also praised it for clarity, accuracy, and ability to follow instructions, according to OpenAI.

“OpenAI o3-pro has access to tools that make ChatGPT useful—it can search the web, analyse files, reason about visual inputs, use Python, personalise responses using memory, and more,” said OpenAI in a post on X.

Sam Altman, CEO, OpenAI said that the model is really smart. “I didn’t believe the win rates relative to o3 the first time I saw them,” said Altman in a post on X.

What does o3-pro model bring?

Feature-wise, o3-pro comes equipped with web search, Python code execution, file analysis, image understanding, and personalised responses via memory. However, it currently can’t generate images, doesn’t support OpenAI’s Canvas workspace feature, and lacks temporary chat support in ChatGPT due to a technical issue. Another trade-off: responses are noticeably slower compared to o1-pro.

Performance-wise, o3-pro appears to lead the pack. It reportedly beat 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, said OpenAI.