MDGym - Easy: leaderboard
MDGym (Kumar et al., 2026) asks an AI agent to set up, run and analyse expert-curated molecular dynamics simulations in LAMMPS or GROMACS and report physical quantities; a task is solved only when every required quantity is within 5% of the reference. Easy tier (57 tasks): full success rate in percent, higher is better, with each entry a coding-agent harness and model pair.
Metric: Full success rate (%; share of the 57 easy expert-curated LAMMPS and GROMACS simulation tasks with every required output quantity within 5% of the reference). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 21.1 | #158 |
| 2 | GPT-5.2 Codex | 21.1 | #185 |
| 3 | Qwen3 Coder | 1.8 | #584 |
| 4 | GPT-OSS-20B | 0 | #1150 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mdgym-easy · How It Works · Data refreshed daily, snapshot 2026-11-12.