OpenAI Launches ChatGPT Specialized in 'Inference'

The inference-capable ChatGPT developed by OpenAI has been unveiled. On the 12th (local time), OpenAI announced the release of a new version of ChatGPT focused on inference capabilities, called 'OpenAI o1 (hereafter O1)'. Until now, OpenAI had been developing O1 under the codename 'Strawberry'.

O1 is an AI model specialized in reasoning through tasks related to mathematics and coding. According to OpenAI, when O1 solved the preliminary problems of the International Mathematical Olympiad (IMO), it achieved an accuracy rate of 83%, overwhelmingly higher than the previous model's 13% accuracy. OpenAI also added that O1 could assist physicists in creating complex mathematical formulas and support medical researchers' experiments.

Jakub Pachocki, OpenAI's chief scientist, said, "Previous models like ChatGPT start responding immediately when asked a question, but this model may take time," adding, "It thinks about the problem in English, analyzes it, finds angles, and presents the best solution."

This new model is part of OpenAI's efforts to develop human-level AI, known as Artificial General Intelligence (AGI). OpenAI divides AI capability levels into five stages and sets goals to develop AGI, the ultimate objective.

Subscribers of ChatGPT Plus and Chat Teams services, as well as enterprises, can use this new technology starting today. OpenAI explained, "Complex reasoning tasks represent significant progress and indicate a new level of AI capability," adding, "Considering this, we reset the counter to 1 and named this model 'OpenAI o1'."