OpenAI releases o1, its first model with ‘reasoning’ abilities

[Logesh-Murugan]

OpenAI officially launched the o1 series of AI models on September 12, 2024. This series includes the o1-preview and o1-mini models, which are designed to enhance reasoning capabilities in complex tasks, particularly in science, coding, and mathematics. Users of ChatGPT Plus and Team can access these models starting from the launch date, with plans for broader access in the future.

The o1 models are trained to spend more time thinking before responding, mimicking human-like problem-solving. This approach allows them to tackle complex challenges more effectively than previous models. For instance, in tests, o1 solved 83% of problems in a mathematics competition, compared to just 13% by earlier models.

The o1 models excel in scientific reasoning, ranking in the 89th percentile on competitive programming questions and demonstrating PhD-level accuracy in physics, biology, and chemistry benchmarks. Enhanced safety protocols have been implemented, making o1 models better at adhering to guidelines and preventing misuse compared to previous iterations.