DiscoverPhysics: leaderboard
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. We introduce DiscoverPhysics, an interactive benchmark that asks a LLM agent to discover the laws of motion of a simulated world whose physics deliberately deviates from our own. We construct 22 worlds governed by, among others, screened
Metric: Mean Explanation Score (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 61 | #49 |
| 2 | GPT-5.5 | 59 | #31 |
| 3 | Qwen 3.5 397B A17B | 33 | #146 |
| 4 | GPT-5.4 | 29 | #79 |
| 5 | Kimi K2.5 | 29 | #144 |
| 6 | Claude Sonnet 4.6 | 28 | #90 |
| 7 | Qwen 3 235B A22B | 24 | #313 |
| 8 | Claude Haiku 4.5 | 23 | #280 |
| 9 | GPT-OSS-120B | 22 | #341 |
| 10 | GPT-OSS-20B | 14 | #518 |
| 11 | Llama 3.3 70B | 10 | #579 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=discoverphysics · How It Works · Data refreshed daily, snapshot 2026-10-20.