—
scBench: A Benchmark for Single-Cell RNA-seq Analysis
Evaluates whether models can solve practical single-cell RNA-seq analysis tasks with deterministic grading. Tasks require empirical interaction with .h5ad data files - agents must load and analyze the data to produce correct answers. Covers 30 canonical tasks across 5 sequencing
single-celltranscriptomicstabularcode
Status
curated
Saturation
unsaturated
Runnable
no
Cost to run
high
Editor's note
Practical single-cell RNA-seq analysis tasks — directly in the dbverse / spatial-omics wheelhouse and a high-value target for a bio-specialist leaderboard. Eight frontier models evaluated; Claude Opus 4.6 leads at 52.8% accuracy and no model clears 53%.
Results
| Model | Score | Variant | Provenance |
|---|---|---|---|
| Claude Opus 4.6 | 52.8% accuracy | default | paper |
| Claude Opus 4.5 | 49.9% accuracy | default | paper |
| GPT-5.2 | 45.2% accuracy | default | paper |
| Claude Sonnet 4.5 | 44.2% accuracy | default | paper |
| GPT-5.1 | 37.9% accuracy | default | paper |
| Grok-4.1 | 35.6% accuracy | default | paper |
| Grok-4 | 33.9% accuracy | default | paper |
| Gemini 2.5 Pro | 29.2% accuracy | default | paper |