o1 (AI model)

2024: Product Release

In mid-September 2024, OpenAI released a new o1 AI model, which the developers say shows excellent results in complex reasoning, outperforming people in math, coding and science tests.

In the qualifying exams of the International Mathematical Olympiad (IMO), the new model correctly solved 83% of the problems, while its predecessor GPT-4o gave only 13% of the correct answers. In Codeforces coding competitions, the model reached the 89th percentile, and the developers believe that it can also be among the top 500 students in the country according to the results of the elite American Mathematics Exam (AIME). In addition, OpenAI claims that when solving complex control problems in physics, chemistry and biology, the model demonstrates results comparable to those of graduate students.

OpenAI

According to the developers, the new AI model is based on a reinforcement learning system designed to teach the model how to solve complex problems using a chain of reasoning. OpenAI claims that the o1 model has developed superior reasoning skills by mimicking the step-by-step logic inherent in humans, correcting errors, and adjusting strategies before issuing a definitive response.

A preview of this AI model, codenamed Project Strawberry, is now available to ChatGPT Pro subscribers as well as through the OpenAI API. OpenAI also introduced the o1-mini, a stripped-down but faster model that developers say is particularly effective at encoding. It is 80% cheaper than o1-preview, which positions it as a more affordable option for applications that require reasoning but do not need broad knowledge of the world. Adding o1 capabilities to ChatGPT in pilots will test the new model within realistic use cases.^[1]

Notes