Atlas

Benchmarks

← All benchmarks

SciUniverse Level 1: Materials

Science

The 23 materials-science tasks of SciUniverse Level 1 in four families: alpha-alumina conversion, BaTiO3 pellet pressing and PVA solution preparation in the physical lab, and quantitative phase analysis of 20 measured XRD patterns. Pass@1 weights the four families equally.

Top models (higher is better)

ModelScore
Claude Fable 5.168.8
Claude Opus 568.1
Grok 4.640.9
GPT-6 Astra20.6
GPT-5.6 Sol15.3
Gemini 3.8 Flash14.7
Loading Atlas data…