AI Solves Long-Standing Mathematical Problems

OpenAI's next-generation model Astra solved 10 mathematical problems that had remained open for at least a decade, at a computational cost of approximately $2,000. CEO Sam Altman personally traveled to Washington to brief the US government on the model's release. Separately, Tencent Hunyuan's research agent Hyra, powered by the Hy3 model, solved a 50-year-old problem in additive combinatorics: determining the optimal exponent for how much faster the sumset |A+A| can grow compared to the difference set |A-A| for a finite set of integers. A 1969 theorem had established an upper bound of 2, but the best constructions for over 50 years barely exceeded 1.1. Hyra found an explicit construction proving the optimal exponent is exactly 2. The result includes a published paper, a blog post, and a formal proof.
Key insights
AI Is Solving Open Mathematical Problems at Trivial Computational Cost
Astra solved 10 decade-old problems for $2,000, and Hyra solved a 50-year-old problem. These results suggest that AI is becoming a cost-effective research tool for pure mathematics, capable of making contributions that would be considered significant achievements for human mathematicians.
Formal Verification Accompanies AI-Generated Proofs
Tencent Hunyuan's result includes a formal proof alongside the paper, indicating that AI-generated mathematical breakthroughs are being validated through mechanized verification, increasing confidence in their correctness.