ProteinGeoBench - Main 130 Tasks: leaderboard
ProteinGeoBench (2026): Harbor executable-verifier mean reward over 130 protein tasks measured for all four agents, from a 131-task suite; medium reasoning requested, default temperature, up to eight runs per task (ten for Gemini). Internet and structure tools allowed; task-balanced mean, not perfect-solve rate. Mean Reward (0–1); higher is better.
Metric: Mean Reward (0–1). Source: arxiv.org. Saturation forecast: Around January 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 (Medium) | 0.82 | #12 (Claude Opus 5) |
| 2 | GPT-5.6 Sol (Medium) | 0.81 | #18 (GPT-5.6 Sol) |
| 3 | Grok 4.5 (Medium) | 0.79 | #47 (Grok 4.5) |
| 4 | Gemini 3.1 Pro (Preview) (Medium) | 0.71 | #58 (Gemini 3.1 Pro (Preview)) |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=proteingeobench-main-130-tasks · How It Works · Data refreshed daily, snapshot 2026-10-21.