scBench (Mini-SWE-Agent) - Cell Typing: leaderboard
Metric: Accuracy (%) on the 118 cell typing problems of scBench; each scBench problem gives an agent a single-cell data snapshot (AnnData) and a natural-language analysis task, and a deterministic grader passes or fails its JSON answer; mini-SWE-agent harness (bash actions, up to 100 steps, 600-second limit), three replicates per problem averaged; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 48.2 | #60 |
| 2 | Claude Opus 4.5 | 45.8 | #79 |
| 3 | GPT-5.2 | 39.5 | #105 |
| 4 | Claude Sonnet 4.5 | 35.6 | #138 |
| 5 | Grok 4.1 | 30.2 | #218 |
| 6 | Grok 4 | 29.1 | #169 |
| 7 | GPT-5.1 | 29.1 | #131 |
| 8 | Gemini 2.5 Pro | 22 | #145 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=scbench-mini-swe-agent-cell-typing · How It Works · Data refreshed daily, snapshot 2026-10-11.