Atlas

Benchmarks

← All benchmarks

Finance Agent v2.1 - All Pass

Professional Work · 2026-10-19

Finance Agent v2.1 strict All-Pass metric over the 449-question test split: every check on a question must pass.

Top models (higher is better)

ModelScore
Gemini 4 Argon58.9
Muse Wonder Full54.9
Gemini 3.8 Flash54.5
Muse Wonder53.1
GLM-5.351.9
Claude Haiku 5.551.8
DeepSeek V4.1 Flash51.7
DeepSeek V4.1 Flash 101751.3
MiMo-V2.6-Flash50.7
MiMo-V2.6-Pro50.4
Grok 4.749.2
Hy4 Preview49.0
Muse Spark Open49.0
GPT-6.1 Sol48.2
GLM-5.3 Flash47.8
Mistral Large 4 Preview47.8
GPT-6 Astra47.5
Kimi K347.3
MiniMax M3.1 Flash45.6
Step 5 Preview44.7
GPT-6 Luna43.8
MiniMax M342.5
DeepSeek V4 Pro 081342.1
Mercury 338.5
Beam 1.0 501B A23B37.9
Inkling36.3
Inkling-Small31.9
Loading Atlas data…