A suite of 50 unpublished research-level theoretical physics problems with verifiable answers.
Without AI assistance, a typical problem could be solved by a world expert in about a day, but would take considerably longer for experts in adjacent specialties.
With realistic tools — literature search and scripting for numerical and symbolic work — the best-performing agent, GPT-6-Astra harnessed in Codex, obtained final answers with near-perfect 98% accuracy at an average cost under USD 3 per problem.