Atlas

Benchmarks

← All benchmarks

HLAB-Vals-1.1 - Criteria Pass Rate

Professional Work · 2026-10-19

HLAB-Vals-1.1 criteria pass rate under Harvey's definition: each judge's fraction of passed criteria, averaged over the two judges. Version 1 required both judges to pass a criterion, so this reads higher and is not comparable with hlab-criteria-pass-rate.

Top models (higher is better)

ModelScore
Gemini 4 Argon96.3
Muse Wonder Full96.1
Muse Wonder96.0
GLM-5.393.7
Grok 4.793.7
Kimi K393.5
Mistral Large 4 Preview93.0
MiMo-V2.6-Pro92.9
MiMo-V2.6-Flash92.8
DeepSeek V4.1 Flash 101792.7
Gemini 3.8 Flash92.6
DeepSeek V4 Pro 081391.9
GPT-6 Astra91.7
GLM-5.3 Flash91.6
GPT-6.1 Sol91.5
MiniMax M3.1 Flash91.1
Muse Spark Open90.1
Claude Haiku 5.590.1
Claude Opus 5.589.8
Mercury 389.6
Hy4 Preview89.6
Beam 1.0 501B A23B89.1
GPT-6 Luna88.3
Claude Sonnet 5.585.9
Inkling-Small81.5
Inkling72.8
Loading Atlas data…