Acropolis

Tax Agent

Tax Agent, a component of the Vals Index.

Upstream source
FinanceAgentic
Independent. Run by an operator independent of the model vendors.Vals AI benchmarks by Vals AIVersion 2.2Data as of 17 Nov 2026

No filters. Showing 38 of 38 rows.

  1. 1
    GoogleDossier

    xhigh

    Accuracy

    Gemini 4 Hydrogen, Accuracy: 52.3% (scale 0.0 to 100.0), 95% confidence interval 45.2 to 59.452.3%(45.2–59.4)
    Duration per test
    1,094.5 s
    Cost per test
    $3.87
  2. 2
    GoogleDossier

    high

    Tied #1

    Accuracy

    Gemini 4 Carbon, Accuracy: 51.8% (scale 0.0 to 100.0), 95% confidence interval 44.7 to 58.851.8%(44.7–58.8)
    Duration per test
    2,086.2 s
    Cost per test
    $6.34
  3. 3
    AlibabaDossier
    Tied #1

    Accuracy

    Qwen3.8 Max Internal, Accuracy: 49.7% (scale 0.0 to 100.0), 95% confidence interval 42.6 to 56.749.7%(42.6–56.7)
    Duration per test
    5,231.2 s
    Cost per test
    $4.07
  4. 4
    GoogleDossier

    high

    Tied #1

    Accuracy

    Gemini 4 Argon, Accuracy: 49.2% (scale 0.0 to 100.0), 95% confidence interval 42.2 to 56.349.2%(42.2–56.3)
    Duration per test
    985.6 s
    Cost per test
    $3.47
  5. 5
    AnthropicDossier

    max

    Tied #1

    Accuracy

    Opus 5.5, Accuracy: 44.5% (scale 0.0 to 100.0), 95% confidence interval 37.5 to 51.644.5%(37.5–51.6)
    Duration per test
    4,012.5 s
    Cost per test
    $14.40
  6. 6
    DeepSeekDossier

    max

    Tied #1

    Accuracy

    DeepSeek V4.1 Flash 1017, Accuracy: 44.4% (scale 0.0 to 100.0), 95% confidence interval 37.3 to 51.444.4%(37.3–51.4)
    Duration per test
    1,264.4 s
    Cost per test
    $0.46
  7. 7

    max

    Tied #1

    Accuracy

    GLM-5.3, Accuracy: 43.3% (scale 0.0 to 100.0), 95% confidence interval 36.3 to 50.343.3%(36.3–50.3)
    Duration per test
    3,052.9 s
    Cost per test
    $1.80
  8. 8

    max

    Tied #1

    Accuracy

    Muse Wonder Full 1.0, Accuracy: 43% (scale 0.0 to 100.0), 95% confidence interval 36.0 to 50.043%(36.0–50.0)
    Duration per test
    1,084.3 s
    Cost per test
    $2.15
  9. 9
    AnthropicDossier

    max

    Accuracy

    Sonnet 5.5, Accuracy: 42% (scale 0.0 to 100.0), 95% confidence interval 35.0 to 48.942%(35.0–48.9)
    Duration per test
    3,560.5 s
    Cost per test
    $8.73
  10. 10
    AlibabaDossier

    xhigh

    Accuracy

    Qwen4 Max, Accuracy: 40.4% (scale 0.0 to 100.0), 95% confidence interval 33.4 to 47.340.4%(33.4–47.3)
    Duration per test
    1,472.7 s
    Cost per test
    $1.03
  11. 11
    Thinking MachinesDossier

    Accuracy

    Inkling 2 Small, Accuracy: 38.9% (scale 0.0 to 100.0), 95% confidence interval 32.5 to 45.338.9%(32.5–45.3)
    Duration per test
    414 s
    Cost per test
    $0.14
  12. 12

    max

    Accuracy

    Muse Ember, Accuracy: 38.2% (scale 0.0 to 100.0), 95% confidence interval 31.4 to 45.138.2%(31.4–45.1)
    Duration per test
    90.1 s
    Cost per test
    $0.032
  13. 13

    max

    Accuracy

    Muse Wonder 1.0, Accuracy: 36.3% (scale 0.0 to 100.0), 95% confidence interval 29.5 to 43.136.3%(29.5–43.1)
    Duration per test
    492.8 s
    Cost per test
    $1.88
  14. 14

    xhigh

    Accuracy

    Grok 4.7, Accuracy: 35.8% (scale 0.0 to 100.0), 95% confidence interval 29.0 to 42.535.8%(29.0–42.5)
    Duration per test
    996.5 s
    Cost per test
    $1.80
  15. 15
    TencentDossier

    Accuracy

    Hy4, Accuracy: 34.4% (scale 0.0 to 100.0), 95% confidence interval 27.9 to 40.934.4%(27.9–40.9)
    Duration per test
    3,807.7 s
    Cost per test
    $1.69
  16. 16
    Moonshot AIDossier

    max

    Accuracy

    Kimi K3, Accuracy: 33.4% (scale 0.0 to 100.0), 95% confidence interval 26.7 to 40.033.4%(26.7–40.0)
    Duration per test
    2,415.2 s
    Cost per test
    $2.79
  17. 17
    TencentDossier

    Accuracy

    Hy4 Preview, Accuracy: 32.6% (scale 0.0 to 100.0), 95% confidence interval 26.0 to 39.232.6%(26.0–39.2)
    Duration per test
    2,464 s
    Cost per test
    $0.59
  18. 18
    XiaomiDossier

    Accuracy

    MiMo-V2.6-Pro, Accuracy: 32.4% (scale 0.0 to 100.0), 95% confidence interval 25.8 to 39.032.4%(25.8–39.0)
    Duration per test
    1,547.7 s
    Cost per test
    $0.13
  19. 19
    MiniMaxDossier

    max

    Accuracy

    MiniMax M3.1 Flash, Accuracy: 32.1% (scale 0.0 to 100.0), 95% confidence interval 25.6 to 38.732.1%(25.6–38.7)
    Duration per test
    1,962.8 s
    Cost per test
    $0.32
  20. 20
    GoogleDossier

    high

    Accuracy

    Gemini 3.8 Flash, Accuracy: 32.1% (scale 0.0 to 100.0), 95% confidence interval 25.5 to 38.732.1%(25.5–38.7)
    Duration per test
    178.3 s
    Cost per test
    $0.86
  21. 21
    AnthropicDossier

    high

    Accuracy

    Fable 5.2, Accuracy: 32.1% (scale 0.0 to 100.0), 95% confidence interval 25.5 to 38.732.1%(25.5–38.7)
    Duration per test
    299.7 s
    Cost per test
    $3.31
  22. 22
    AnthropicDossier

    max

    Accuracy

    Haiku 5.5, Accuracy: 31.9% (scale 0.0 to 100.0), 95% confidence interval 25.4 to 38.531.9%(25.4–38.5)
    Duration per test
    2,973.6 s
    Cost per test
    $1.16
  23. 23

    max

    Accuracy

    GLM-5.3 Flash, Accuracy: 31.6% (scale 0.0 to 100.0), 95% confidence interval 25.0 to 38.131.6%(25.0–38.1)
    Duration per test
    1,297 s
    Cost per test
    $0.14
  24. 24
    MistralDossier

    high

    Accuracy

    Mistral Large 4 OSS, Accuracy: 30.7% (scale 0.0 to 100.0), 95% confidence interval 24.2 to 37.130.7%(24.2–37.1)
    Duration per test
    1,023.3 s
    Cost per test
    $0.50
  25. 25
    XiaomiDossier

    Accuracy

    MiMo-V2.6-Flash, Accuracy: 28.5% (scale 0.0 to 100.0), 95% confidence interval 22.4 to 34.728.5%(22.4–34.7)
    Duration per test
    1,038.9 s
    Cost per test
    $0.049
  26. 26
    StepFunDossier

    Accuracy

    Step 5 Preview, Accuracy: 27.3% (scale 0.0 to 100.0), 95% confidence interval 21.2 to 33.427.3%(21.2–33.4)
    Duration per test
    1,288.5 s
    Cost per test
    $0.55
  27. 27
    DeepSeekDossier

    max

    Accuracy

    DeepSeek V4 Pro 0813, Accuracy: 26.8% (scale 0.0 to 100.0), 95% confidence interval 20.7 to 32.826.8%(20.7–32.8)
    Duration per test
    1,596.8 s
    Cost per test
    $0.73
  28. 28

    xhigh

    Accuracy

    Muse Spark Open, Accuracy: 26.3% (scale 0.0 to 100.0), 95% confidence interval 21.1 to 31.526.3%(21.1–31.5)
    Duration per test
    701.8 s
    Cost per test
    $0.88
  29. 29
    Moonshot AIDossier

    max

    Accuracy

    Kimi K3.1 Pro, Accuracy: 24.9% (scale 0.0 to 100.0), 95% confidence interval 18.8 to 31.024.9%(18.8–31.0)
    Duration per test
    3,518.6 s
    Cost per test
    $3.64
  30. 30
    OpenAIDossier

    max

    Accuracy

    GPT-6.1 Sol, Accuracy: 21.8% (scale 0.0 to 100.0), 95% confidence interval 15.9 to 27.621.8%(15.9–27.6)
    Duration per test
    1,213.9 s
    Cost per test
    $1.32
  31. 31
    OpenAIDossier

    max

    Accuracy

    GPT-6 Astra, Accuracy: 20.7% (scale 0.0 to 100.0), 95% confidence interval 15.0 to 26.420.7%(15.0–26.4)
    Duration per test
    900.6 s
    Cost per test
    $5.80
  32. 32
    OpenAIDossier

    max

    Accuracy

    GPT-6 Luna, Accuracy: 19.9% (scale 0.0 to 100.0), 95% confidence interval 14.3 to 25.619.9%(14.3–25.6)
    Duration per test
    1,995.7 s
    Cost per test
    $0.19
  33. 33
    InceptionDossier

    high

    Accuracy

    Mercury 3, Accuracy: 18.2% (scale 0.0 to 100.0), 95% confidence interval 13.2 to 23.118.2%(13.2–23.1)
    Duration per test
    269.3 s
    Cost per test
    $0.046
  34. 34
    OpenAIDossier

    max

    Accuracy

    GPT-6.1 Terra, Accuracy: 17.1% (scale 0.0 to 100.0), 95% confidence interval 11.8 to 22.417.1%(11.8–22.4)
    Duration per test
    2,751.4 s
    Cost per test
    $2.16
  35. 35
    Thinking MachinesDossier

    Accuracy

    Inkling Small, Accuracy: 15.4% (scale 0.0 to 100.0), 95% confidence interval 10.8 to 20.015.4%(10.8–20.0)
    Duration per test
    385.3 s
    Cost per test
    $0.092
  36. 36
    Thinking MachinesDossier

    Accuracy

    Inkling, Accuracy: 15.3% (scale 0.0 to 100.0), 95% confidence interval 10.5 to 20.015.3%(10.5–20.0)
    Duration per test
    810.6 s
    Cost per test
    $0.35
  37. 37
    NVIDIADossier

    high

    Accuracy

    Nemotron 3.5 Super VL, Accuracy: 11.6% (scale 0.0 to 100.0), 95% confidence interval 7.3 to 15.911.6%(7.3–15.9)
    Duration per test
    72.7 s
    Cost per test
    $0.046
  38. 38
    ReflectionDossier

    Accuracy

    Beam 1.0, Accuracy: 8.8% (scale 0.0 to 100.0), 95% confidence interval 5.1 to 12.58.8%(5.1–12.5)
    Duration per test
    58 s
    Cost per test
    Not tested