CompBioBench v1 - Author Baselines: leaderboard
CompBioBench v1 (Genentech): 100 agentic computational-biology tasks across genomics, transcriptomics, epigenomics, single-cell analysis, human genetics and ML workflows, each with a single exact-match answer. This series keeps only the benchmark authors' baseline runs, each model in its developer's CLI, averaged over their submissions; accuracy %, higher is better.
Metric: Accuracy (%, 100 agentic computational-biology tasks, mean of author runs). Source: huggingface.co. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 83.3 | #83 |
| 2 | Gemini 3.1 Pro (Preview) | 82 | #64 |
| 3 | Claude Opus 4.6 | 80.3 | #67 |
| 4 | Claude Opus 4.7 | 78.3 | #51 |
| 5 | Claude Sonnet 4.6 | 70 | #94 |
| 6 | Gemini 3 Flash | 66 | #103 |
| 7 | Claude Haiku 4.5 | 34 | #287 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=compbiobench-v1-author-baselines · How It Works · Data refreshed daily, snapshot 2026-11-05.