Artificial intelligence is experiencing a qualitative leap in the field of pure mathematics. AIbase learned that OpenAI's GPT-5.2Pro model recently successfully assisted in solving problem number 281 in the field of number theory, the "Erdős problem." Renowned mathematician Terence Tao gave it high praise, calling it "one of the most explicit cases" of AI solving open mathematical problems.

According to the information, this breakthrough was driven by researcher Neel Somani. Although previous related proofs may have provided background references for the model, after comparison, Tao confirmed that the proof process given by GPT-5.2Pro was "completely different" from previous methods. AIbase noticed that this was not the first attempt of the model; as early as January 4, 2026, the model had already achieved an autonomous solution to the Erdős problem.

However, while celebrating, the mathematical community remains clear-headed. Tao warned that public perception of AI's capabilities might be biased. Since failed attempts are rarely published, people often only see successful cases. AIbase learned that a tracking database established by Paata Ivanisvili and Mehmet Mars Seven shows that the actual success rate of AI attempting to solve such problems is only 1% to 2%, and mostly concentrated on easier questions. Despite this, the potential of AI as a research tool cannot be ignored.

Key points:

  • 🧠 Breakthrough: GPT-5.2Pro successfully solved the number theory problem #281, with an original proof logic that has been recognized by the mathematical master Terence Tao.

  • 📊 Truth about success rate: The latest database reveals that the real success rate of AI solving such problems is only about 1%-2%, and it still struggles with medium to high difficulty complex mathematical challenges.

  • 🛠️ New research tool: Although not a panacea, AI has been proven to provide problem-solving approaches different from traditional human paths, becoming a powerful aid in mathematical research.