Atlas

Benchmarks

← All benchmarks

Diligence Stack Agent — Source Grounding

Professional Work · 2026-07-09

Diligence Stack Agent Source Grounding is the normalized percentage of available rubric points earned for relevant evidence, attribution, and traceability, averaged across evaluated work.

Top models (higher is better)

ModelScore
Sonnet 591.5
GPT-5.6 Luna90.8
GPT-5.6 Sol90.2
Claude Fable 589.1
GPT-5.588.3
Muse Spark 1.182.5
GPT-5.6 Terra82.0
Opus 4.880.8
Grok 4.580.0
Kimi K380.0
Grok 4.365.0
Gemini 3.5 Flash58.8
DeepSeek-V4-Pro57.8
Kimi K2.655.0
Loading Atlas data…