OpenAI has introduced o3-pro, a new AI model the company describes as its most capable yet. The model builds on o3, the reasoning model OpenAI launched earlier this year, and is now being placed in front of both ChatGPT users and developers through the API.
The release is aimed at users who need stronger step-by-step reasoning in areas where accuracy and structure matter. OpenAI says expert reviewers preferred o3-pro over o3 across every tested category, including science, education, programming, business, and writing help.
A stronger version of o3
O3-pro is not a separate product category. It is a version of OpenAI's o3, a reasoning model designed to work through problems step by step instead of responding in the more direct pattern used by conventional AI models.
That distinction matters because OpenAI positions reasoning models as more reliable in domains such as physics, math, and coding. These are areas where a model often needs to follow a chain of logic, handle constraints, and avoid skipping important intermediate steps.
OpenAI says o3-pro uses the same underlying model as o3. Because of that, the company points users to the o3 system card for full safety details.
Who gets access and when
O3-pro is available in ChatGPT for Pro and Team users starting Tuesday. In the model picker, it replaces OpenAI o1-pro, which means users in those plans will see o3-pro take the position previously held by that older premium reasoning option.
Enterprise and Edu users will get access the week after, according to OpenAI. The staggered rollout means the model reaches individual and team subscribers first, then expands to institutional customers.
Developers also get access through OpenAI's developer API. The model went live in the API as of this afternoon, giving builders a way to put o3-pro into applications and workflows outside ChatGPT.
API pricing and token basics
OpenAI has priced o3-pro at $20 per million input tokens and $80 per million output tokens in the API. Input tokens are the tokens fed into the model. Output tokens are the tokens generated by the model in response.
The source article notes that a million input tokens is equivalent to about 750,000 words. That comparison helps put the pricing unit into plain language: API costs are tied to the amount of material sent to the model and the amount of material it produces.
For developers, the price structure makes both sides of a request important. A long prompt, a large file, or a detailed instruction set affects input tokens, while a long answer affects output tokens.
Tools, strengths, and slower responses
O3-pro has access to tools, according to OpenAI. That allows it to search the web, analyze files, reason about visual inputs, use Python, personalize its responses using memory, and more.
Those capabilities make the model broader than a text-only assistant. It can work with files, visual material, code execution, and web information when those tools are available in the product environment.
OpenAI also says expert reviewers rated o3-pro higher than o3 for clarity, comprehensiveness, instruction-following, and accuracy. The company specifically identified science, education, programming, business, and writing help as key domains where reviewers preferred the new model.
The tradeoff is speed. According to OpenAI, o3-pro responses typically take longer than o1-pro to complete. That makes the model more suited to tasks where quality and reasoning depth matter more than getting the fastest possible response.
Limits users should know
O3-pro arrives with several limitations. Temporary chats with the model in ChatGPT are disabled for now while OpenAI resolves a "technical issue." The model also cannot generate images.
Canvas, OpenAI's AI-powered workspace feature, is not supported by o3-pro. That means users who rely on Canvas will not be able to use this model inside that feature at launch.
OpenAI is also pointing to benchmark performance as a strength. In the company's internal testing, o3-pro scores better than Google's top-performing AI model, Gemini 2.5 Pro, on AIME 2024, which evaluates math skills. It also beats Anthropic's recently released Claude 4 Opus on GPQA Diamond, a test of PhD-level science knowledge.
Taken together, the release shows OpenAI pushing its reasoning model lineup toward more capable, tool-using systems. For ChatGPT users, o3-pro replaces o1-pro in supported plans. For developers, it adds a higher-priced API option built around step-by-step reasoning, tool access, and stronger performance in demanding tasks.