GPT-6 Astra Solves Complex Physics Benchmarks for Just $3

15 Oct 2026, 7:10 pm · 1h ago · 1 min read · OfficeChaiNew

A new theoretical physics evaluation benchmark, TheoPhysEval, developed by a team of nearly 30 physicists, reveals that OpenAI's GPT-6 Astra solved 49 out of 50 unpublished physics research problems. The benchmark assessed the model's capacity to tackle complex, novel theoretical physics problems that were not previously available on the public internet, demonstrating advanced analytical reasoning capabilities at an estimated computing cost of $3 per problem. This milestone indicates rapid progress in generative AI models moving beyond basic mathematics into cutting-edge scientific research and automated scientific discovery.