AInsteinBench - Numerical Relativity: leaderboard
The AInsteinBench problems from Einstein Toolkit and AMReX only (the paper's NR column; the subset size is shown only in a figure), solved by the model in the CodeAct agent; the score is the share resolved. ByteDance Seed, which also makes Doubao-Seed-1.8, wrote the benchmark and ran every model.
Metric: % Resolved. Source: arxiv.org. Saturation forecast: Around January 2028. 7 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 45.8 | #120 | |
| 2 | Claude Opus 4.5 | 44.1 | #161 | |
| 3 | Claude Sonnet 4.5 | 37.3 | #258 | |
| 4 | GLM-4.6 | 37.3 | #496 | |
| 5 | GPT-5.1 | 35.6 | #235 | |
| 6 | GPT-5 | High | 35.6 | #177 |
| 7 | Seed 1.8 | 34.5 | #250 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=ainsteinbench-numerical-relativity · How It Works · Data refreshed daily, snapshot 2026-11-17.