OpenAI released the new generation AI inference models o3 and o3-mini on Friday. These two models are currently in the testing phase and will be gradually launched after January next year.

This move highlights the intense competition in the AI large model market. Currently, competitors like Google are racing to release their own AI models aiming to solve some complex problems.

OpenAI CEO Sam Altman stated that OpenAI plans to launch o3 mini by the end of January 2025, after which the full version of o3 will be released. The new and more powerful AI models help attract new investments and users.

In September this year, OpenAI launched the inference models o1 and o1-mini for the first time, designed to spend more time processing queries and reasoning to solve some challenges. The main difference from the mainstream model GPT-4o is its ability to solve complex problems such as programming and mathematics while also explaining its reasoning process.

OpenAI claims that the o1 model can reason through complex tasks and can tackle more challenging problems than previous scientific, coding, and mathematical models. The new o3 and o3-mini models will be more powerful than the earlier o1 model. Currently, these two models are undergoing internal safety testing.

OpenAI also stated that the company is opening an application for external researchers to test it before the official release of o3. The event's end time is January 10.

Regarding the o3 model, OpenAI CEO Altman boldly stated that it opens up the 'next phase' of AI. In the ARC-AGI benchmark test, o3 achieved scores ranging from 75.7% to 87.5%, surpassing the human level benchmark of 85%.

The performance of o3 is so outstanding that some people immediately question whether OpenAI has really achieved the holy grail of the AI Industry: Artificial General Intelligence (AGI). Ultraman mentioned earlier this month that AGI could be realized as soon as next year.

The release of the o3 model also marks the last day of OpenAI's 12-day online product launch event. Additionally, OpenAI did not release the o2 model, but instead jumped directly to o3.