OpenAI’s latest model, GPT-5.2 Pro, has achieved a milestone in AI mathematics by scoring 31% on the challenging FrontierMath benchmark’s Tier 4, surpassing Gemini 3 Pro’s previous high of 19% and GPT-5.2 xhigh’s 17%. Epoch AI conducted these tests manually on the ChatGPT platform due to API constraints. Notably, GPT-5.2 Pro successfully solved 15 out of 48 tasks, including four previously unsolved problems. While feedback from mathematicians has been largely favorable, some experts criticized the model’s occasional imprecision in explanations. These benchmark results align with other reports indicating that AI models, particularly GPT-5 variants, are significantly improving in solving complex mathematical challenges, including Erdős problems. However, mathematician Terence Tao advises caution against drawing hasty conclusions about the model’s capabilities. Stay updated on AI advancements and mathematics solutions by subscribing for curated news.
Source link
