Openai has launched the O3-Pro, the AI model the company claims to be the most capable of ever.
The O3-Pro is the O3 version of Openai, a reasoning model launched by a startup earlier this year. In contrast to traditional AI models, inference models work staged problems, allowing them to run more reliably in domains such as physics, mathematics, coding, and more.
The O3-Pro will be available to CHATGPT Pro and team users starting Tuesday and will replace the O1-PRO model. Enterprise and EDU users will be able to access it in a week, Openai says. O3-Pro will also be live on Openai’s developer API this afternoon.
The O3-Pro priced is $20 per million input tokens and $80 per million API’s $1 million. An input token is a token that is fed to the model, and an output token is a token that the model generates based on the input token.
A million input tokens equals about 750,000 words, slightly longer than “war and peace.”
“In expert evaluations, reviewers always prefer O3-Pro over O3-Pro in all tested categories, especially major domains such as science, education, programming, business, writing help,” Openai writes in Changelog. “The reviewers also consistently praised the O3-Pro for clarity, inclusiveness, follow-up instruction and accuracy.”
According to Openai, O3-Pro has access to tools, allowing you to search the web, analyze files, personalize reasons for visual input, use Python, memory-enabled responses, and more. The downside is that according to OpenAI, the response of the model is usually longer than the O1-Pro.
O3-Pro has other limitations. Temporary chat with ChatGPT models is currently disabled, but Openai solves “technical issues.” O3-Pro cannot generate images. Openai’s AI-powered workspace feature Canvas is not supported by O3-Pro.
On the plus side, according to Openai’s internal testing, the O3-Pro has achieved an impressive score on the popular AI benchmark. At AIME 2024, which evaluates model mathematical skills, Gemini 2.5 Pro is Google’s top-performing AI model. The O3-Pro also defeats the recently released Claude 4 Opus on GPQA Diamond, a test of PHD-level scientific knowledge.
Source link
