OpenAI has introduced a groundbreaking series of models called ‘o1’ that specialize in advanced reasoning. These models, previously referred to as Strawberry AI, include OpenAI o1, OpenAI o1-preview, and OpenAI o1-mini. Starting today, paid ChatGPT Plus users can access the preview and mini versions, with plans to extend OpenAI o1-mini to free users in the near future. The key highlight of these models is their ability to excel at solving complex problems in fields like math, science, and coding. OpenAI explains that the o1 models take a moment to process before delivering responses, but this added thinking time allows them to handle intricate tasks with greater accuracy.
According to OpenAI, the o1 models perform on par with PhD students when tackling difficult scientific topics. In a recent benchmark, the OpenAI o1 model achieved an impressive 83% score in the International Mathematics Olympiad (IMO) exam, a significant leap from GPT-4o’s 13% performance. The o1 model also demonstrated its superior coding abilities, placing in the 89th percentile in the highly competitive Codeforces programming competition, compared to GPT-4o’s 11th percentile ranking.
These developments represent a significant leap in AI’s capacity for sophisticated thinking and problem-solving. With the advancement of AI models, OpenAI’s o1 series has the potential to revolutionize the field by enabling users to take on increasingly complex tasks in several fields. By granting both premium and free users access to these potent models, OpenAI keeps expanding the realm of what artificial intelligence is capable of.