BioDesignBench (Guided Mode): leaderboard
The 76 BioDesignBench protein design tasks (Kim and Romero, 2026) run in guided mode: the same tools are grouped by function and common tool sequences are packaged as composite workflows with usage hints, so the agent is shown the pipeline instead of working it out. Designs are scored out of 100 by computed structural and physical metrics plus a panel of model judges; higher is better. Unguided runs of the same tasks are the BioDesignBench series.
Metric: Overall Score. Source: huggingface.co. Saturation forecast: Around January 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 | 58.46 | #705 |
| 2 | GPT-5 | 55.26 | #163 |
| 3 | Claude Sonnet 4.5 | 50.23 | #244 |
| 4 | Gemini 2.5 Pro | 8.75 | #262 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=biodesignbench-guided-mode · How It Works · Data refreshed daily, snapshot 2026-11-12.