Atlas

Benchmarks

← All benchmarks

AML Bench pass^3

Professional Work · 2026-10-19

Share of AML Bench's 153 simulated anti-money-laundering investigations that an agent passes on all three of its attempts, from the same runs as the AML Bench pass@1. An attempt passes only if every grading criterion on the filed disposition passes.

Top models (higher is better)

ModelScore
GPT-6.1 Sol66.7
Claude Opus 5.564.7
Claude Sonnet 5.564.7
GLM-5.358.2
Kimi K357.5
GLM-5.3 Flash56.9
Gemini 4 Argon54.9
Grok 4.753.6
DeepSeek V4.1 Flash52.9
GPT-5.6 Terra48.4
Gemini 3.8 Flash42.5
Loading Atlas data…