scBench: leaderboard
Bioinformatics agent benchmark with verifiable single-cell RNA-seq workflow tasks and deterministic graders.
Metric: Accuracy (self-reported). Source: benchmarklist.com. Status: saturation imminent. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Mythos 5 | 59.3 |
| 2 | Claude Opus 4.8 | 58.2 |
| 3 | Claude Mythos Preview | 58.2 |
| 4 | GPT-5.5 | 57.95 |
| 5 | GPT-5.4 | 57.44 |
| 6 | Claude Opus 4.7 | 55.3 |
| 7 | Gemini 3.1 Pro (Preview) | 53.85 |
| 8 | Claude Opus 4.6 | 52.65 |
| 9 | GPT-5.2 | 52.31 |
| 10 | Claude Sonnet 4.6 | 50.4 |
| 11 | Claude Opus 4.5 | 47.18 |
| 12 | Grok 4.20 Beta (0309) (Reasoning) | 44.44 |
| 13 | Grok 4.3 | 44.27 |
| 14 | GPT-5.1 | 38.8 |
| 15 | Claude Sonnet 4.5 | 33.16 |
Interactive version: theaggregate.ai/benchmark?slug=scbench · How It Works · Data refreshed daily, snapshot 2026-09-05.