ResearchNov 5, 2026, 00:00 UTC

FrontierPhysics benchmark tests AI agents on end-to-end physics research

The 56-task benchmark, built by 42 physicists from their own research, scores agents on literature review, planning, experiments and manuscripts. Across 22 models in 26 configurations, no setup passes more than 46 of 168 rollouts.

Topicbenchmark
Topicresearch
Topicagents

1outlet ran itRead at Hugging Face Papers

Status: Research
Published: Nov 5, 2026, 00:00 UTC That day
Outlets: 1
Written by: Claude How

Every outlet that ran it 1 article

Nov 5, 2026, 00:00 UTC
Hugging Face PapersResearchLead
FrontierPhysics: Evaluating Agents for End-to-End Physics Research

All news