Atlas

Benchmarks

← All benchmarks

Vals Finance Index

Indexes · 2026-10-21

Vals Finance Index v1 weights three agentic finance benchmarks run by Vals by each sub-sector's share of U.S. GDP: (1.6 x the mean of Finance Agent v2.1 and EMB v1.1, plus 0.5 x Tax Agent Bench's citation-adjusted all-pass rate) / 2.1. Finance Agent v2.1 covers 449 analyst research questions, EMB v1.1 102 financial modeling tasks and Tax Agent Bench 193 corporate tax research tasks. Component runs match the standalone Vals leaderboards, which Atlas imports separately; mixed-model fallback aggregates are excluded from single-model observations. Tracked for comparison only; as an index it is not a capability-model input.

Top models (higher is better)

ModelScore
Gemini 4 Argon64.9
Muse Wonder Full64.6
Muse Wonder61.2
DeepSeek V4.1 Flash 101760.9
GLM-5.360.5
Gemini 3.8 Flash60.5
Grok 4.758.5
Kimi K358.0
MiMo-V2.6-Pro57.8
Hy4 Preview57.5
Claude Haiku 5.557.4
MiMo-V2.6-Flash57.1
DeepSeek V4.1 Flash57.1
Mistral Large 4 Preview56.9
MiniMax M3.1 Flash55.9
GLM-5.3 Flash55.8
GPT-6.1 Sol55.8
GPT-6 Astra55.6
Step 5 Preview55.4
GPT-6 Luna53.1
Muse Spark Open53.0
DeepSeek V4 Pro 081352.9
MiniMax M348.4
Mercury 347.2
Beam 1.0 501B A23B46.5
Inkling38.2
Inkling-Small36.3
Loading Atlas data…