DeepFinance

An ultra long-horizon benchmark for measuring models on real-world Private Equity and Investment Banking workflows that would take a human professional between 20 and 30 hours to complete.

tasks : 21

average human time : 27.1 hours per task

rubric criteria per task : 84

files per data room : 19

Mean rubric score (%)

0%10%20%30%40%50%60%70%

Julius Research

  1. 54.8

    Claude Opus 5.5(max)

  2. 54.6

    Claude Sonnet 5.5(max)

  3. 54.2

    GPT-6 Astra(xhigh)

  4. 52.9

    GPT-6.1 Sol(xhigh)

  5. 45.6

    Claude Fable 5.1(max)

  6. 43.5

    GPT-6.1 Terra(xhigh)

  7. 41.3

    Gemini 4 Argon(high)

  8. 37.0

    Muse Spark 1.3(max)

  9. 34.6

    Kimi K3.1 Pro(max)

  10. 34.1

    Grok 4.7(xhigh)

  11. 30.3

    Qwen 3.8 Max(xhigh)

  12. 29.8

    Inkling 2 Small(max)

  13. 27.7

    GLM 5.3(max)

  14. 27.3

    DeepSeek V4.1 Flash(max)

  15. 24.6

    GPT-6 Luna(xhigh)

  16. 21.5

    Mistral Large 4(high)

  17. 20.5

    DeepSeek V4 Pro(max)

  18. 14.5

    Nemotron 3 Ultra(high)

  19. 13.1

    Inkling(max)

  20. 10.9

    Inkling Small(max)