AssayBench - Precision@100: leaderboard
Share of an AssayBench model's top 100 ranked genes that the screen actually called hits, on the temporal split. Higher is better. Source: genentech.github.io/AssayBench.
Metric: Precision@100 (%). Source: genentech.github.io. Saturation forecast: Around 2033. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 16.67 | #82 |
| 2 | Gemini 3 Flash | 14.34 | #103 |
| 3 | Gemini 3.1 Pro (Preview) | 14.22 | #64 |
| 4 | Claude Sonnet 4.5 | 14.02 | #148 |
| 5 | GPT-5.2 | 13.57 | #115 |
| 6 | GPT-5 Mini | 13.39 | #193 |
| 7 | Claude Opus 4.5 | 12.84 | #87 |
| 8 | GPT-5.4 | 12.84 | #83 |
| 9 | GLM-5 | 11.88 | #149 |
| 10 | Qwen 3 235B A22B 2507 | 11.65 | #341 |
| 11 | Kimi K2.5 | 11.59 | #152 |
| 12 | Qwen 3.5 397B A17B | 11.47 | #154 |
| 13 | GPT-OSS-20B | 11.05 | #523 |
| 14 | Qwen 3.5 122B A10B | 11.04 | #184 |
| 15 | Claude Haiku 4.5 | 10.68 | #287 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=assaybench-precision-100 · How It Works · Data refreshed daily, snapshot 2026-11-05.