AssayBench - Adjusted nDCG@100 (Random Split): leaderboard
AssayBench hit ranking on the random split, the easier control for the temporal split, over the same LLMs. Adjusted nDCG at 100, higher is better. Source: genentech.github.io/AssayBench.
Metric: Adjusted nDCG@100 (%). Source: genentech.github.io. Saturation forecast: Around 2030. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 16.02 | #82 |
| 2 | Gemini 3.1 Pro (Preview) | 15.1 | #64 |
| 3 | Gemini 3 Flash | 15.07 | #103 |
| 4 | GPT-5.4 | 14.69 | #83 |
| 5 | GPT-5.2 | 13.88 | #115 |
| 6 | GPT-5 Mini | 13.79 | #193 |
| 7 | Claude Opus 4.5 | 13.5 | #87 |
| 8 | Claude Sonnet 4.5 | 13.31 | #148 |
| 9 | Kimi K2.5 | 12.77 | #152 |
| 10 | GLM-5 | 12.71 | #149 |
| 11 | Qwen 3.5 397B A17B | 11.63 | #154 |
| 12 | DeepSeek V3.2 | 11.32 | #210 |
| 13 | GPT-OSS-20B | 11.26 | #523 |
| 14 | Qwen 3.5 35B A3B | 10.99 | #268 |
| 15 | Qwen 3.5 122B A10B | 10.95 | #184 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=assaybench-adjusted-ndcg-100-random-split · How It Works · Data refreshed daily, snapshot 2026-11-05.