OpenAI unveils new o3 and o3-mini rational AIs; check this out

In a week full of announcements, OpenAI unveiled its newest “rational” language models this Friday (20). o3 and o3-mini. The technologies have only been briefly presented by the company with some mathematical resolution and programming tests, but still appear to be far from being publicly available.

According to OpenAI tests, the o3 outperforms its predecessor by almost 23% in programming tests using SWE-Bench Verified. The model also achieved 87% accuracy when solving the GPQA Diamond, a complex advanced math test that not all technologies can solve.

Additionally, in a competitive pure programming scenario, OpenAI’s new feature was 44% better than the existing o1 the company uses.

Is rational artificial intelligence the future?

The announcement of o3 and o3-mini may confuse those who follow the innovations of OpenAI and other companies in the world of artificial intelligence. In September, the company introduced the o1 and o1-mini models with the same rational offer, and about four months passed between the introductions.

Rational AIs were developed for more advanced uses, such as solving math problems and assisting professionals with programming. The difference with the resource is that it was created to think in a human-like way, improving its judgment to create different intervention strategies and learn from mistakes.

If you want to test the new o3 and o3-mini, you’d better wait. It will take time for the models to be released and there is no date yet, but OpenAI will open tests with researchers and developers who want to use the resources in their projects.