LABBench2
Science · 2026-02-04
Edison Scientific's successor to LAB-Bench measures practical biology research using PDFs, images, databases and bioinformatics artifacts. Google's reported aggregate gives models a Linux terminal with bioinformatics tools, Python, R and internet access; the exact evaluated denominator and aggregate weighting are not disclosed.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 88.8 |
| Gemini 3.8 Flash | 86.2 |
| GPT-6 Astra | 85.4 |
| Gemini 3.7 Flash | 82.1 |
| Claude Opus 5.5 | 73.1 |
| Claude Fable 5.1 | 68.6 |