BioDesignBench: leaderboard
76 protein design tasks covering antibodies, enzymes, mini binders, scaffolds and fluorescent proteins, each either designed from scratch or redesigned from an existing sequence. An agent plans and calls tools such as RFdiffusion, ProteinMPNN and AlphaFold2 over repeated rounds, then submits a design scored out of 100 by computed structural and physical metrics plus a panel of model judges. These are the unguided runs, in which each tool comes with only its own description and the agent must work out the pipeline itself; runs with workflow guidance are the separate BioDesignBench (Guided Mode) series. Higher is better.
Metric: Overall Score. Source: huggingface.co. Saturation forecast: Around 2031. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 | 60.43 | #321 |
| 2 | GPT-5 | 55.61 | #98 |
| 3 | Claude Sonnet 4.5 | 41.17 | #142 |
| 4 | Gemini 2.5 Pro | 8.11 | #152 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=biodesignbench · How It Works · Data refreshed daily, snapshot 2026-10-20.