On the final day of the 12-day long ‘Shipmas’, OpenAI revealed a new set of frontier reasoning models called O3 and O3-mini. The models are only previewed, and OpenAI won’t be releasing them anytime soon.
OpenAI CEO Sam Altman said the plan is to launch the O3 Mini by the end of January and the full O3 shortly after. This is because Altman believes that stronger LLMs can outperform existing models and attract new investors and users.
However, the AI startup is accepting applications from researchers to test these models before making them public. The application process for external researchers will close on January 10.
Interestingly, OpenAI launched O1, previously code-named Project Strawberry, in September this year. Now, it looks like the company is ditching the o2 altogether, and moving forward with the o3 models. According to reports, the company is leaving o2 to avoid a possible copyright conflict with a British company called O2.
According to reports, the o3 model has surpassed the previous performance record set by its predecessors. The model outperformed the coding tests, called SWE-Bench verified, by 22.8 percent. The model reportedly outscored OpenAI’s chief scientist in competitive programming. In addition, the model also scored 87.7 percent on expert-level science problems in AIME 2024, one of the toughest math competitions. When it comes to difficult math and logic problems, the o3 solved 25.02 percent of them, beating other models that couldn’t score more than 2 percent.
While the o1 models are capable of reasoning through complex tasks and solving difficult problems, the new o3 and o3-mini models will reportedly be even more powerful. The new models are currently undergoing internal safety testing.
Why should you buy our membership?
You want to be the smartest in the room.
You want access to our award-winning journalism.
You don’t want to be confused and misinformed.
Choose your subscription package