All benchmarks

Finance Agent v2

Agentic financial analysis

Finance Agent evaluates models on realistic financial-analysis workflows — pulling figures from filings and market data, running calculations and producing well-reasoned analyses under real-world constraints. Version 2 is 927 expert-reviewed questions across nine analytical categories, run by Vals AI with EDGAR search, web search, document parsing and retrieval tools.

Model scores

  • Fable 5.158.9%
  • Fable 556.3%
  • Opus 5.558.6%
  • Opus 558.6%
  • Opus 4.853.9%
  • Sonnet 5.558.1%
  • Sonnet 553.9%
  • GPT-6 Astra53.5%
  • GPT-6.1 Sol52.0%
  • GPT-5.6 Sol53.8%
  • Grok 4.752.3%
  • GPT-5.551.8%
  • Composer 2.5—
  • Opus 4.751.5%
  • Gemini 3.1 Pro43.0%
  • Mythos Preview—

Official source: Finance Agent v2 leaderboard (Vals AI)

Related reading