Skip to content
Fifth Order

Research

Perspectives and evidence at the intersection of finance and AI.

Explore benchmarks

Financial Workflows Benchmark

Evaluates source-grounded, multi-step professional finance work: gathering evidence from filings and transcripts, reconciling figures across documents, and completing analysis with an inspectable trail. Completion is assessed step by step—retrieval, calculation, synthesis and judgment—rather than by a final answer alone, because professional workflows fail at intermediate steps. Higher scores indicate more completed, correctly evidenced steps.

RankModelScore
  1. 1
    xAIGrok 4.6 xHigh reasoning
    88.2%
  2. 2
    OpenAIGPT 5.6 Luna Max reasoning
    87.2%
  3. 3
    ZhipuGLM 5.3 Max reasoning
    82.2%