GPT-6-Astra solves 49 of 50 physics research problems at $2.40 apiece
AISummary
A physics benchmark of 50 routine research problems tested commercially available AI agents without human hints.
GPT-6-Astra, running in OpenAI's Codex, solved 49 of 50 problems, or 98 percent, at under $3 per problem and just under $116 for the whole benchmark.
The authors say the problems resemble first research projects for PhD students, and that the models may have memorized much of the theoretical physics literature.
AIWhy it matters
The source shows which models solved routine theoretical physics problems and where they still fail, helping readers gauge how far AI research agents have progressed.
Source: The Decoder · the-decoder.comPublished · added here