SC-Arena - Cell Generation: leaderboard
Metric: Knowledge-augmented judge score from 0 to 100 on SC-Arena cell generation (write a plausible cell sentence for 608 cell types); a GPT-4o-mini judge given the reference answer and curated external knowledge (Cell Ontology, CellMarker, Gene Ontology, UniProt, NCBI and source literature) rates each answer 0 to 5, rescaled to 0 to 100 and averaged over the task's items; cells are written as cell sentences of ranked gene names; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2 | 63.04 | #236 |
| 2 | DeepSeek R1 | 62.24 | #245 |
| 3 | GPT-4o | 59.7 | #333 |
| 4 | Qwen 3 32B | 55.39 | #424 |
| 5 | Qwen 3 235B A22B | 52.76 | #304 |
| 6 | Qwen 2.5 14B | 51.74 | #711 |
| 7 | Qwen 2.5 32B | 50.95 | #532 |
| 8 | Qwen 3 8B | 50.53 | #667 |
| 9 | Qwen 2.5 7B | 45.98 | #912 |
| 10 | Qwen 3 14B | 15.5 | #524 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=sc-arena-cell-generation · How It Works · Data refreshed daily, snapshot 2026-10-11.