OpenAI could unveil its most powerful AI model yet on the final day of the 12-day OpenAI extravaganza. The new model is a next-generation “reasoner” that will replace the o1 announced last week.

The Information reports that it could be called o3, skipping o2, the name of the British telecom company. This was corroborated by OpenAI CEO Sam Altman's cryptic post on X, “I should have said oh oh oh.”

We know nothing about this new model, but some experts suggest that it can solve evaluation tests designed to determine if a model is artificial intelligence (AGI).

If the model is declared AGI (which Altman says will be achieved in 2025), OpenAI could terminate its long-standing contract with Microsoft and renegotiate more favorable terms with the technology giant.

Whether this is a big leap for AI or a small step for ChatGPT will become clear at 1 pm EST (6 pm CST, 5 am EST).

This new announcement may be a major step forward in AI inference, an area where OpenAI's o-series is already making headway. But even in that area, it faces competition from Google with its new Gemini Flash Thinking model.

OpenAI's o-series models operate slightly differently from GPT families such as GPT-4o. For example, o1 focuses on reasoning and problem solving, using chains of thoughts to solve problems. In contrast, GPT-4o is trained to process and generate multiple modalities through a unified neural network.

Part of the difference in o1 is that it has switched to focus on posterior training; OpenAI uses tools to improve the way it processes specific tasks and learns from problems. This is known as reinforcement learning and has been confirmed to be available to developers on the second day of the 12-Days event.

Altman had previously hinted at the integration of a feature that appears to be the rumored “Orion” project. It is not clear whether today's announcement is that itself with a new name, or whether o3 is simply an improved version of o1.

We seem to be skipping o2 altogether because of the UK telco of the same name. o2 recently announced an AI project that uses generated voice to keep scammers from calling virtual old ladies.