Benchmark Radar
AI BENCHMARK PROFILE

DiscoverPhysics

General AIAgents

DiscoverPhysics is an interactive benchmark for LLM agents to discover laws of motion in simulated worlds with altered physics, scoring trajectory MSE and explanation quality.

Released
2026-05-25
Readiness
Paper only
Primary field
General AI

Why it matters

It probes long-horizon reasoning and hypothesis refinement, distinguishing recall from genuine scientific reasoning.

Motivation

Frontier LLMs now perform strongly across a wide range of physics evaluations, but it is hard to disentangle genuine reasoning from recall of established science.

Primary resources

Benchmark Radar records only publicly supported details and links back to primary sources for verification.