Atlas

Benchmarks

← All benchmarks

SciUniverse Level 1

Science

C5R's 92 tasks in 17 families of short scientific work across chemistry, biology and materials science, from sample preparation, instrument control and protocol adaptation to reaction optimization, facility management and interpreting real NMR, XRD and chromatography data. Models direct Facility-0, a physical lab, or its digital twin through C5R's facility harness, controlling instruments and instructing human operators. Pass@1 averages repeats within tasks and weights every task family equally.

Top models (higher is better)

ModelScore
Claude Fable 5.145.3
GPT-6 Astra32.5
Claude Opus 530.5
Grok 4.626.2
Gemini 3.8 Flash14.6
GPT-5.6 Sol9.4
Loading Atlas data…