OpenAI has presented its most powerful AI models - o3 and o3-mini. These new products turned out to be more advanced compared to the previous line of o1 models. They were named with the omission of the number 2, due to possible legal problems with the British operator O2, which registered the brand. OpenAI chief executive Sam Altman said the choice of names follows the company's tradition. According to OpenAI, the o3 model performed outstandingly well in the ARC benchmark, scoring 87.5% - well above the human average (around 85%) and much better than the o1, which only achieved 32%. However, the cost of performing tasks on the new model could be around $2,000. The o3 also performed well on a special math exam in the US, scoring 96.7 points and failing only one question. In the Frontier Math benchmark, the model set a record score of 25.2%, while other models failed to score more than 2%. The results of the o3-mini were more modest, but OpenAI is confident that it outperforms the o1. At the moment, both models are not available to a wide audience. Developers are continuing the process of training them and collecting applications from researchers to test o3 and o3-mini. OpenAI has not yet announced a timeline for a full release