TwinLab-Bench - Mean Score: leaderboard
TwinLab-Bench (arXiv 2611.26248) scientific experimentation benchmark (Mean Score) across 52 tasks in agentic twins of real laboratories.
Metric: Mean Rubric Score (0-100 across 52 tasks). Source: huggingface.co. Saturation forecast: Around May 2028. 13 models tracked. Updated 22 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 | High | 45 | #2 |
| 2 | Claude Opus 5.5 | High | 40.2 | #9 |
| 3 | GPT-6.1 Sol | High | 31.7 | #15 |
| 4 | GPT-6 Astra | High | 30.7 | #11 |
| 5 | Claude Sonnet 5.5 | High | 28.7 | #24 |
| 6 | DeepSeek V4.1 Flash (1017) | High | 20 | #48 |
| 7 | GLM-5.3 | High | 19.4 | #79 |
| 8 | Kimi K3.1 Pro | High | 18.1 | #35 |
| 9 | MiMo-V2.6-Pro | High | 17.9 | #94 |
| 10 | Qwen 3.8 Max | High | 13.1 | #67 |
| 11 | hy4 | High | 12.9 | #44 |
| 12 | MiMo-V2.6-Flash | High | 10.9 | #183 |
| 13 | GPT-6 Luna | High | 10.6 | #108 |
No result here: #3 Gemini 4 Neon, #4 GPT-6.2 Astra, #7 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=twinlab-bench-mean-score · How It Works · Data refreshed daily, snapshot 2026-11-22.