Openai comes out O3-Pro, an inflated version of its o3 ai reasoning model

OPENAI A spear O3-Pro, an AI model that society claims is the most capable to date.

O3-Pro is a version of Openai O3A model of reasoning that the startup launched earlier this year. Unlike conventional AI models, reasoning models are resistant to step -by -step problems, allowing them to operate more reliable in fields such as physics, mathematics and coding.

O3-Pro is available for Chatgpt Pro and team users from Tuesday, replacing the O1-Pro model. Users of the company and EDU will have access the following week, explains Openai. O3-Pro has also been live in the OpenAi developer API since this afternoon.

O3-Pro is for $ 20 per million entry tokens and $ 80 per million exit tokens in the API. The entry tokens are tokens supplied in the model, while the output tokens are tokens that the model generates according to the entry tokens.

A million entry tokens is equivalent to around 750,000 words, a little longer than “war and peace”.

“In expert assessments, examiners systematically prefer O3-Pro compared to O3 in each category tested and especially in key fields such as science, education, programming, business and writing aid,” writes Openai in a modification. “The examiners have also evaluated the O3-Pro constantly higher for more clarity, understanding, monitoring of instructions and precision.”

O3-Pro has access to tools, according to Openai, allowing it to search for the web, to analyze files, reason on visual entries, to use Python, to personalize your answers by pulling memory, etc. As an inconvenience, the model’s responses generally take longer than O1-Pro to finish, according to Openai.

O3-Pro has other limits. Temporary cats with the model in Chatgpt are deactivated for the moment while Openai solves a “technical problem”. O3-Pro cannot generate images. And Canvas, the workspace function powered by OPENAI, is not supported by O3-Pro.

On the positive side, O3-Pro gets impressive scores in popular benchmarks, according to the internal tests of Openai. On Aime 2024, which assesses the skills in mathematics of a model, O3-Pro score better than the most efficient AI model of Google, Gemini 2.5 Pro. O3-Pro also beats Anthropic recently released Close the 4th work On GPQA Diamond, a knowledge of science at doctorate.