Issue 2026-10-15 · Industry · 研究 · 应用

GPT-6-Astra solves 49 of 50 physics problems at $2.40 each

GPT-6-Astra solves 49 of 50 physics problems at $2.40 each

On the TheoPhysEval benchmark, GPT-6-Astra solved 49 of 50 unpublished theoretical physics calculations as accurately as leading specialists, at $2.40 per problem. In some apparent failures, experts' reference solutions were wrong. Researchers say physics hasn't seen math-like breakthroughs because agents lack 'taste,' but expect autonomous top-level AI research soon.

The Decoder17 d ago
Read original ↗