last updated 2026-10-19

AML Bench

An agent benchmark for anti-money-laundering investigation. The agent is the second-level investigator on a case handed over by a colleague: it reads the ledgers, re-screens the parties, applies the bank’s written procedure and files a disposition. The grade is the state it leaves the case system in, never what it says in chat.

Leaderboard

Agentpass@1every criterion95% CIpass³all 3 repeatsSingle-turn (105)everything is on fileMulti-turn (48)must ask the colleagueAskedon multi-turn tasks$/taskHarness
GPT-6.1 Sol70.6%63.6–77.366.7%90.5%27.1%8.3%~$0.26Codex
Sonnet 5.566.9%59.7–74.164.7%88.9%18.8%0.0%~$0.21Claude Code
Opus 5.566.9%59.5–74.164.7%83.5%30.6%12.5%~$0.44Claude Code
GLM 5.366.4%59.5–73.258.2%88.3%18.8%0.0%~$0.14Pi
GLM 5.3 Flash66.4%59.7–73.256.9%88.6%18.1%0.0%≈$0.02Pi
Kimi K365.6%58.6–72.357.5%87.3%18.1%0.0%~$0.31Pi
Grok 4.763.8%57.1–70.653.6%85.1%17.4%0.0%~$0.46Grok CLI
Gemini 4 Argon63.4%56.4–70.454.9%79.4%28.5%7.6%~$0.47Gemini CLI
DeepSeek V4.1 Flash61.7%54.5–68.652.9%81.9%17.4%0.0%~$0.06Pi
GPT-5.6 Terra61.0%54.0–67.848.4%80.6%18.1%0.0%~$0.29Codex
Gemini 3.8 Flash56.9%50.1–63.642.5%76.2%14.6%0.0%~$0.42Gemini CLI
Nemotron 3.5 Super VL54.5%47.9–61.034.0%71.4%17.4%0.0%~$0.05Hermes

Cost vs. cases passed

50%55%60%65%70%75%$0.02$0.05$0.10$0.20$0.50Mean agent-side cost per case (USD, log scale) · Y: pass@1 (% of attempts)Nemotron 3.5 Super VLGemini 3.8 FlashGPT-5.6 TerraDeepSeek V4.1 FlashGemini 4 ArgonGrok 4.7Kimi K3GLM 5.3GLM 5.3 FlashSonnet 5.5Opus 5.5GPT-6.1 SolGPT-6.1 SolSonnet 5.5Opus 5.5GLM 5.3GLM 5.3 FlashKimi K3Grok 4.7Gemini 4 ArgonDeepSeek V4.1 FlashGPT-5.6 TerraGemini 3.8 FlashNemotron 3.5 Super

Unlock the full AML Bench