FrontierPhysics Benchmark Tests AI Research Agents with Physicist-Designed Tasks

A new benchmark called FrontierPhysics, organized by physicists from Berkeley, Caltech and Harvard, features 56 research tasks designed by 42 physicists across nine areas of physics to evaluate AI research agents end-to-end. Claude Opus 5.5 topped the leaderboard with 27.4%.

2026-11-05 ~ 2026-11-06 · 2 related posts

• FrontierPhysics: 56 expert tasks benchmark AI agents on end-to-end physics research — Bingran You · 2026-11-05
• FrontierPhysics benchmark: Claude Opus 5.5 tops end-to-end physics research at 27.4% — xeophon · 2026-11-06