Atlas

Benchmarks

← All benchmarks

FrontierPhysics v0.1 Verifier Pass

Science

Share of a configuration's 168 FrontierPhysics v0.1 rollouts (56 physics research tasks, three trials each) whose deliverables pass the task's code verifier, without the further checks of the headline Overall Pass. It comes from the same rollouts as the headline and is a component of its factor group.

Top models (higher is better)

ModelScore
Claude Opus 5.540.5
GPT-6 Astra34.5
Gemini 4 Argon33.3
GPT-6.1 Sol31.0
Claude Fable 5.128.6
GPT-6 Sol21.4
GPT-5.6 Sol17.9
Grok 4.717.9
Claude Opus 516.7
GPT-5.6 Terra14.9
Claude Haiku 5.514.3
GPT-5.6 Luna11.9
MiMo-V2.6-Pro9.5
Gemini 3.8 Flash7.7
GLM-5.37.1
Mistral Large 4 Preview4.8
Grok 4.64.2
Kimi K33.0
Loading Atlas data…