OpenAI has introduced o3-pro, an AI model the company asserts is its most advanced
O3-pro is an enhanced variant of OpenAI’s o3, a reasoning model the company introduced earlier this year. In contrast to traditional AI models, reasoning models solve problems incrementally, which allows them to operate more consistently in fields such as physics, mathematics, and coding.
Starting Tuesday, the o1-pro model will be replaced by the O3-pro variant for ChatGPT Pro and Team users. According to OpenAI, access will be granted to enterprise and educational users the following week. O3-pro is also operational in OpenAI’s developer API as of this afternoon.
In the API, O3-pro is priced at $20 per million input tokens and $80 per million output tokens. Output tokens are tokens that the model generates based on the input tokens, while input tokens are those given into the model.
About 750,000 words are equivalent to one million input tokens, slightly lengthier than “War and Peace.”
In a changelog, OpenAI states that “in expert evaluations, reviewers consistently prefer o3-pro over o3 in every tested category, especially in key domains like science, education, programming, business, and writing help.” “O3-pro was consistently rated higher by reviewers regarding clarity, comprehensiveness, accuracy, and instruction-following.”
According to OpenAI, O3-pro can search the web, analyze files, reason about visual inputs, use Python, personalize its responses by leveraging memory, and more. According to OpenAI, the model’s responses typically require more time to complete than those of o1-pro, which is a disadvantage.
O3-pro is subject to additional constraints. While OpenAI resolves a “technical issue,” temporary conversations with the model in ChatGPT are currently disabled. O3-pro is incapable of producing images. Additionally, o3-pro does not support Canvas, OpenAI’s AI-powered workspace feature.
Additionally, OpenAI’s internal testing indicates that o3-pro accomplishes remarkable results in prevalent AI benchmarks. The o3-pro model outperforms Google’s top-performing AI model, Gemini 2.5 Pro, on the AIME 2024 test, which assesses a model’s mathematical abilities.
O3-pro also surpasses Anthropic’s recently released Claude 4 Opus on the GPQA Diamond, a test that evaluates the scientific knowledge of individuals with a PhD.
