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

A group of researchers built a benchmark of 50 unpublished theoretical physics problems with verifiable answers and tested AI agents on them without hints. GPT-6-Astra, running in OpenAI's Codex environment, solved 49 of 50 at under $3 each, matching specialists. The authors say agents still lack the judgment, or taste, behind real breakthroughs.

Summary written with AI by HowTechMoves, based on reporting by The Decoder.