Atlas

Benchmarks

← All benchmarks

Immigration Bench - Intake Checks

Professional Work · 2026-10-28

Vals' Intake Checks panel of Immigration Bench, apparently the rubric items asking the model to confirm facts about the person's situation (the page's payload counts intake checks per question by persona; international-student questions have none). The page does not define the panel or its denominator, and its standard errors are zero placeholders. Diagnostic only.

Top models (higher is better)

ModelScore
Claude Fable 5.158.9
Claude Opus 5.549.7
Muse Wonder Full48.3
Claude Haiku 5.547.7
Claude Sonnet 5.547.7
GLM-5.347.7
Muse Wonder47.7
MiMo-V2.6-Pro45.6
Gemini 4 Argon45.0
DeepSeek V4 Pro 081343.5
Kimi K343.0
Muse Spark 1.343.0
GPT-6 Astra42.3
Grok 4.741.6
Gemini 3.8 Flash40.9
Grok 4.640.3
DeepSeek V4.1 Flash39.6
GPT-6.1 Sol39.6
Mistral Large 4 Preview39.6
GPT-6 Luna38.9
Muse Spark 1.238.9
MiniMax M336.9
Gemini 3.5 Flash-Lite31.5
Loading Atlas data…