TwinLab-Bench - Pass Rate: leaderboard

TwinLab-Bench (arXiv 2611.26248) scientific experimentation benchmark (Pass Rate) across 52 tasks in agentic twins of real laboratories.

Metric: Task Pass Rate (%, out of 52 tasks). Source: huggingface.co. Saturation forecast: Around May 2028. 13 models tracked. Updated 22 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Fable 5.2High46.15#2
2Claude Opus 5.5High42.31#9
3GPT-6 AstraHigh32.69#11
4GPT-6.1 SolHigh32.69#15
5Claude Sonnet 5.5High30.77#24
6DeepSeek V4.1 Flash (1017)High21.15#48
7GLM-5.3High21.15#79
8Kimi K3.1 ProHigh19.23#35
9MiMo-V2.6-ProHigh19.23#94
10hy4High13.46#44
11Qwen 3.8 MaxHigh13.46#67
12GPT-6 LunaHigh11.54#108
13MiMo-V2.6-FlashHigh11.54#183

No result here: #3 Gemini 4 Neon, #4 GPT-6.2 Astra, #7 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=twinlab-bench-pass-rate · How It Works · Data refreshed daily, snapshot 2026-11-22.