Acropolis

Vals Index

A single measure of AI's potential economic impact: agentic model performance across finance, coding, legal, and tax tasks, weighted by each sector's share of U.S. GDP.

Upstream source
AgenticFinanceCodingLegalComposite of vals-finance-agent, vals-emb, vals-terminal-bench-4, vals-vibe-code-bench, vals-code-migration, vals-legal-research, vals-legal-agent-benchmark, vals-tax-agent-bench
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
    AnthropicDossier

    high

    Accuracy

    Fable 5.2, Accuracy: 71.3% (scale 0.0 to 100.0), 95% confidence interval 69.4 to 73.271.3%(69.4–73.2)
    Duration per test
    2,621.8 s
    Cost per test
    $17.94
    Input tokens per test
    31,113,729.5
    Output tokens per test
    197,828.3
    Reasoning tokens per test
    73,248.9
    Cache-read tokens per test
    30,455,144.1
    Cache-write tokens per test
    658,141
  2. 2
    AnthropicDossier

    max

    Tied #1

    Accuracy

    Sonnet 5.5, Accuracy: 70.9% (scale 0.0 to 100.0), 95% confidence interval 69.0 to 72.770.9%(69.0–72.7)
    Duration per test
    5,158.8 s
    Cost per test
    $16.18
    Input tokens per test
    77,990,774.9
    Output tokens per test
    594,549.6
    Reasoning tokens per test
    405,075.1
    Cache-read tokens per test
    76,487,919.7
    Cache-write tokens per test
    1,502,345.7
  3. 3
    GoogleDossier

    high

    Tied #1

    Accuracy

    Gemini 4 Carbon, Accuracy: 70.5% (scale 0.0 to 100.0), 95% confidence interval 68.5 to 72.470.5%(68.5–72.4)
    Duration per test
    2,864.9 s
    Cost per test
    $10.51
    Input tokens per test
    18,080,265.7
    Output tokens per test
    205,620.3
    Reasoning tokens per test
    93,952.5
    Cache-read tokens per test
    17,241,005
    Cache-write tokens per test
    0
  4. 4
    AnthropicDossier

    max

    Tied #1

    Accuracy

    Opus 5.5, Accuracy: 70.1% (scale 0.0 to 100.0), 95% confidence interval 68.2 to 72.170.1%(68.2–72.1)
    Duration per test
    5,071.1 s
    Cost per test
    $24.77
    Input tokens per test
    59,002,489.2
    Output tokens per test
    455,039.7
    Reasoning tokens per test
    293,955.7
    Cache-read tokens per test
    57,785,923.7
    Cache-write tokens per test
    1,216,167.3
  5. 5
    GoogleDossier

    high

    Tied #1

    Accuracy

    Gemini 4 Argon, Accuracy: 69.4% (scale 0.0 to 100.0), 95% confidence interval 67.3 to 71.569.4%(67.3–71.5)
    Duration per test
    2,698 s
    Cost per test
    $13.37
    Input tokens per test
    23,148,371.5
    Output tokens per test
    159,229.4
    Reasoning tokens per test
    58,309.9
    Cache-read tokens per test
    21,661,266.3
    Cache-write tokens per test
    0
  6. 6
    GoogleDossier

    xhigh

    Accuracy

    Gemini 4 Hydrogen, Accuracy: 68.5% (scale 0.0 to 100.0), 95% confidence interval 66.5 to 70.468.5%(66.5–70.4)
    Duration per test
    1,924.7 s
    Cost per test
    $9.84
    Input tokens per test
    40,855,551.8
    Output tokens per test
    336,848.8
    Reasoning tokens per test
    167,058.3
    Cache-read tokens per test
    39,532,546.3
    Cache-write tokens per test
    0
  7. 7

    max

    Accuracy

    Muse Wonder Full 1.0, Accuracy: 66.8% (scale 0.0 to 100.0), 95% confidence interval 64.7 to 68.966.8%(64.7–68.9)
    Duration per test
    3,558.6 s
    Cost per test
    $6.81
    Input tokens per test
    18,271,139.5
    Output tokens per test
    103,432.9
    Reasoning tokens per test
    48,524.7
    Cache-read tokens per test
    17,923,954.3
    Cache-write tokens per test
    0
  8. 8
    OpenAIDossier

    max

    Accuracy

    GPT-6 Astra, Accuracy: 65% (scale 0.0 to 100.0), 95% confidence interval 62.9 to 67.265%(62.9–67.2)
    Duration per test
    2,018.9 s
    Cost per test
    $21.28
    Input tokens per test
    8,045,900.6
    Output tokens per test
    99,101.8
    Reasoning tokens per test
    51,905.9
    Cache-read tokens per test
    7,645,164.4
    Cache-write tokens per test
    Not tested
  9. 9

    max

    Accuracy

    Muse Wonder 1.0, Accuracy: 64.9% (scale 0.0 to 100.0), 95% confidence interval 62.8 to 67.064.9%(62.8–67.0)
    Duration per test
    2,123.6 s
    Cost per test
    $10.65
    Input tokens per test
    33,073,266.1
    Output tokens per test
    115,811.6
    Reasoning tokens per test
    59,021.7
    Cache-read tokens per test
    32,525,940.1
    Cache-write tokens per test
    0
  10. 10
    OpenAIDossier

    max

    Accuracy

    GPT-6.1 Sol, Accuracy: 63.9% (scale 0.0 to 100.0), 95% confidence interval 61.8 to 66.063.9%(61.8–66.0)
    Duration per test
    2,664 s
    Cost per test
    $3.61
    Input tokens per test
    9,291,986.4
    Output tokens per test
    112,502.2
    Reasoning tokens per test
    62,769
    Cache-read tokens per test
    8,988,645.9
    Cache-write tokens per test
    302,569
  11. 11
    AlibabaDossier

    xhigh

    Accuracy

    Qwen4 Max, Accuracy: 63.2% (scale 0.0 to 100.0), 95% confidence interval 61.2 to 65.263.2%(61.2–65.2)
    Duration per test
    3,879.5 s
    Cost per test
    $9.33
    Input tokens per test
    27,077,683.9
    Output tokens per test
    192,550.1
    Reasoning tokens per test
    95,589.4
    Cache-read tokens per test
    22,483,097.4
    Cache-write tokens per test
    Not tested
  12. 12
    Moonshot AIDossier

    max

    Accuracy

    Kimi K3.1 Pro, Accuracy: 62.5% (scale 0.0 to 100.0), 95% confidence interval 60.3 to 64.862.5%(60.3–64.8)
    Duration per test
    3,816.5 s
    Cost per test
    $6.44
    Input tokens per test
    19,980,586
    Output tokens per test
    159,071.7
    Reasoning tokens per test
    96,642.2
    Cache-read tokens per test
    19,358,617.6
    Cache-write tokens per test
    612,758.3
  13. 13
    DeepSeekDossier

    max

    Accuracy

    DeepSeek V4.1 Flash 1017, Accuracy: 61.3% (scale 0.0 to 100.0), 95% confidence interval 59.2 to 63.361.3%(59.2–63.3)
    Duration per test
    2,394.7 s
    Cost per test
    $0.66
    Input tokens per test
    43,644,461.6
    Output tokens per test
    258,045.4
    Reasoning tokens per test
    174,021.7
    Cache-read tokens per test
    43,300,368.5
    Cache-write tokens per test
    0
  14. 14
    OpenAIDossier

    max

    Accuracy

    GPT-6.1 Terra, Accuracy: 60.5% (scale 0.0 to 100.0), 95% confidence interval 58.6 to 62.560.5%(58.6–62.5)
    Duration per test
    2,383.3 s
    Cost per test
    $2.94
    Input tokens per test
    17,431,997.4
    Output tokens per test
    157,425.5
    Reasoning tokens per test
    110,518.8
    Cache-read tokens per test
    16,948,747.7
    Cache-write tokens per test
    482,147.7
  15. 15

    max

    Accuracy

    Muse Ember, Accuracy: 60.5% (scale 0.0 to 100.0), 95% confidence interval 58.4 to 62.660.5%(58.4–62.6)
    Duration per test
    902.4 s
    Cost per test
    $0.15
    Input tokens per test
    23,054,672.2
    Output tokens per test
    79,012.5
    Reasoning tokens per test
    35,351.4
    Cache-read tokens per test
    22,776,627
    Cache-write tokens per test
    0
  16. 16

    max

    Accuracy

    GLM-5.3, Accuracy: 59.8% (scale 0.0 to 100.0), 95% confidence interval 57.7 to 62.059.8%(57.7–62.0)
    Duration per test
    3,864.1 s
    Cost per test
    $8.52
    Input tokens per test
    27,949,602.7
    Output tokens per test
    202,266.1
    Reasoning tokens per test
    Not tested
    Cache-read tokens per test
    27,410,401.9
    Cache-write tokens per test
    0
  17. 17
    TencentDossier

    Accuracy

    Hy4, Accuracy: 59.4% (scale 0.0 to 100.0), 95% confidence interval 57.2 to 61.559.4%(57.2–61.5)
    Duration per test
    3,091.5 s
    Cost per test
    $1.72
    Input tokens per test
    17,253,359.4
    Output tokens per test
    238,203.3
    Reasoning tokens per test
    180,595.7
    Cache-read tokens per test
    16,708,451.1
    Cache-write tokens per test
    0
  18. 18
    AlibabaDossier

    Accuracy

    Qwen3.8 Max Internal, Accuracy: 59% (scale 0.0 to 100.0), 95% confidence interval 56.9 to 61.259%(56.9–61.2)
    Duration per test
    6,345.6 s
    Cost per test
    $8.73
    Input tokens per test
    26,354,758.4
    Output tokens per test
    231,723.9
    Reasoning tokens per test
    155,010.9
    Cache-read tokens per test
    25,797,460.1
    Cache-write tokens per test
    Not tested
  19. 19
    AnthropicDossier

    max

    Accuracy

    Haiku 5.5, Accuracy: 58.7% (scale 0.0 to 100.0), 95% confidence interval 56.2 to 61.258.7%(56.2–61.2)
    Duration per test
    2,996.8 s
    Cost per test
    $3.17
    Input tokens per test
    37,579,232.6
    Output tokens per test
    570,127.1
    Reasoning tokens per test
    459,156.5
    Cache-read tokens per test
    36,601,039.8
    Cache-write tokens per test
    977,877.9
  20. 20

    xhigh

    Accuracy

    Grok 4.7, Accuracy: 58.6% (scale 0.0 to 100.0), 95% confidence interval 56.6 to 60.658.6%(56.6–60.6)
    Duration per test
    2,013.8 s
    Cost per test
    $11.63
    Input tokens per test
    11,268,193.4
    Output tokens per test
    124,146.2
    Reasoning tokens per test
    68,843.5
    Cache-read tokens per test
    10,616,594.6
    Cache-write tokens per test
    0
  21. 21
    XiaomiDossier

    Accuracy

    MiMo-V2.6-Pro, Accuracy: 57.8% (scale 0.0 to 100.0), 95% confidence interval 55.4 to 60.257.8%(55.4–60.2)
    Duration per test
    4,355.9 s
    Cost per test
    $0.43
    Input tokens per test
    17,623,254.7
    Output tokens per test
    147,276.8
    Reasoning tokens per test
    95,076.7
    Cache-read tokens per test
    17,072,238
    Cache-write tokens per test
    0
  22. 22
    MistralDossier

    high

    Accuracy

    Mistral Large 4 OSS, Accuracy: 56.5% (scale 0.0 to 100.0), 95% confidence interval 54.1 to 59.056.5%(54.1–59.0)
    Duration per test
    2,868 s
    Cost per test
    $3.72
    Input tokens per test
    23,299,961.7
    Output tokens per test
    122,413.2
    Reasoning tokens per test
    Not tested
    Cache-read tokens per test
    20,225,701
    Cache-write tokens per test
    Not tested
  23. 23
    Thinking MachinesDossier

    Accuracy

    Inkling 2 Small, Accuracy: 56.1% (scale 0.0 to 100.0), 95% confidence interval 53.8 to 58.456.1%(53.8–58.4)
    Duration per test
    3,034.5 s
    Cost per test
    $3.51
    Input tokens per test
    108,128,102.6
    Output tokens per test
    281,310.7
    Reasoning tokens per test
    Not tested
    Cache-read tokens per test
    107,436,010.6
    Cache-write tokens per test
    0
  24. 24
    GoogleDossier

    high

    Accuracy

    Gemini 3.8 Flash, Accuracy: 56% (scale 0.0 to 100.0), 95% confidence interval 53.7 to 58.256%(53.7–58.2)
    Duration per test
    2,158.5 s
    Cost per test
    $5.03
    Input tokens per test
    24,930,668.6
    Output tokens per test
    130,206.4
    Reasoning tokens per test
    68,436.7
    Cache-read tokens per test
    23,839,395.3
    Cache-write tokens per test
    0
  25. 25
    XiaomiDossier

    Accuracy

    MiMo-V2.6-Flash, Accuracy: 54.8% (scale 0.0 to 100.0), 95% confidence interval 52.6 to 57.054.8%(52.6–57.0)
    Duration per test
    4,028 s
    Cost per test
    $0.22
    Input tokens per test
    26,689,642.9
    Output tokens per test
    172,166.6
    Reasoning tokens per test
    106,440.1
    Cache-read tokens per test
    25,928,380.5
    Cache-write tokens per test
    0
  26. 26
    StepFunDossier

    Accuracy

    Step 5 Preview, Accuracy: 54.8% (scale 0.0 to 100.0), 95% confidence interval 52.5 to 57.054.8%(52.5–57.0)
    Duration per test
    3,928.9 s
    Cost per test
    $2.73
    Input tokens per test
    33,491,605.9
    Output tokens per test
    159,796.1
    Reasoning tokens per test
    Not tested
    Cache-read tokens per test
    32,751,577.7
    Cache-write tokens per test
    0
  27. 27
    MiniMaxDossier

    max

    Accuracy

    MiniMax M3.1 Flash, Accuracy: 53.8% (scale 0.0 to 100.0), 95% confidence interval 51.7 to 55.953.8%(51.7–55.9)
    Duration per test
    2,597.8 s
    Cost per test
    $0.66
    Input tokens per test
    31,517,784.7
    Output tokens per test
    255,284.6
    Reasoning tokens per test
    183,512.9
    Cache-read tokens per test
    31,161,890.7
    Cache-write tokens per test
    0
  28. 28
    TencentDossier

    Accuracy

    Hy4 Preview, Accuracy: 53.2% (scale 0.0 to 100.0), 95% confidence interval 51.0 to 55.453.2%(51.0–55.4)
    Duration per test
    3,562.6 s
    Cost per test
    $1.57
    Input tokens per test
    19,074,387.8
    Output tokens per test
    160,621
    Reasoning tokens per test
    105,106.1
    Cache-read tokens per test
    18,555,324.8
    Cache-write tokens per test
    0
  29. 29
    DeepSeekDossier

    max

    Accuracy

    DeepSeek V4 Pro 0813, Accuracy: 53% (scale 0.0 to 100.0), 95% confidence interval 50.9 to 55.053%(50.9–55.0)
    Duration per test
    3,482.9 s
    Cost per test
    $2.50
    Input tokens per test
    43,178,249.2
    Output tokens per test
    169,366.1
    Reasoning tokens per test
    97,785
    Cache-read tokens per test
    42,932,509.5
    Cache-write tokens per test
    0
  30. 30
    Moonshot AIDossier

    max

    Accuracy

    Kimi K3, Accuracy: 53% (scale 0.0 to 100.0), 95% confidence interval 51.0 to 54.953%(51.0–54.9)
    Duration per test
    4,496 s
    Cost per test
    $7.17
    Input tokens per test
    15,196,000.1
    Output tokens per test
    107,341.6
    Reasoning tokens per test
    58,602.9
    Cache-read tokens per test
    14,785,760
    Cache-write tokens per test
    285,912.5
  31. 31
    OpenAIDossier

    max

    Accuracy

    GPT-6 Luna, Accuracy: 52.3% (scale 0.0 to 100.0), 95% confidence interval 50.1 to 54.552.3%(50.1–54.5)
    Duration per test
    1,943.2 s
    Cost per test
    $0.45
    Input tokens per test
    17,348,582.2
    Output tokens per test
    170,502.4
    Reasoning tokens per test
    124,778.7
    Cache-read tokens per test
    16,946,836.9
    Cache-write tokens per test
    400,386.5
  32. 32

    max

    Accuracy

    GLM-5.3 Flash, Accuracy: 50.9% (scale 0.0 to 100.0), 95% confidence interval 48.4 to 53.350.9%(48.4–53.3)
    Duration per test
    3,830.6 s
    Cost per test
    $2.05
    Input tokens per test
    62,213,903.2
    Output tokens per test
    203,646.5
    Reasoning tokens per test
    Not tested
    Cache-read tokens per test
    60,767,985.3
    Cache-write tokens per test
    0
  33. 33

    xhigh

    Accuracy

    Muse Spark Open, Accuracy: 48.6% (scale 0.0 to 100.0), 95% confidence interval 46.4 to 50.748.6%(46.4–50.7)
    Duration per test
    4,925.4 s
    Cost per test
    $4.72
    Input tokens per test
    9,735,429.3
    Output tokens per test
    65,357.2
    Reasoning tokens per test
    33,474.7
    Cache-read tokens per test
    6,991,662.4
    Cache-write tokens per test
    0
  34. 34
    InceptionDossier

    high

    Accuracy

    Mercury 3, Accuracy: 43.7% (scale 0.0 to 100.0), 95% confidence interval 41.3 to 46.143.7%(41.3–46.1)
    Duration per test
    1,325.9 s
    Cost per test
    $0.48
    Input tokens per test
    20,204,701.2
    Output tokens per test
    117,494
    Reasoning tokens per test
    105,632.2
    Cache-read tokens per test
    17,311,926.8
    Cache-write tokens per test
    0
  35. 35
    ReflectionDossier

    Accuracy

    Beam 1.0, Accuracy: 43.4% (scale 0.0 to 100.0), 95% confidence interval 41.3 to 45.543.4%(41.3–45.5)
    Duration per test
    446.6 s
    Cost per test
    Not tested
    Input tokens per test
    5,912,885.2
    Output tokens per test
    41,553
    Reasoning tokens per test
    22,329.7
    Cache-read tokens per test
    5,673,911.8
    Cache-write tokens per test
    Not tested
  36. 36
    NVIDIADossier

    high

    Accuracy

    Nemotron 3.5 Super VL, Accuracy: 29.8% (scale 0.0 to 100.0), 95% confidence interval 27.8 to 31.829.8%(27.8–31.8)
    Duration per test
    514 s
    Cost per test
    $0.63
    Input tokens per test
    10,873,912.1
    Output tokens per test
    96,498.1
    Reasoning tokens per test
    57,370.1
    Cache-read tokens per test
    9,860,042.9
    Cache-write tokens per test
    0
  37. 37
    Thinking MachinesDossier

    Accuracy

    Inkling, Accuracy: 29% (scale 0.0 to 100.0), 95% confidence interval 26.8 to 31.229%(26.8–31.2)
    Duration per test
    1,032.9 s
    Cost per test
    $1.85
    Input tokens per test
    7,482,988.3
    Output tokens per test
    64,607.2
    Reasoning tokens per test
    Not tested
    Cache-read tokens per test
    7,298,094.8
    Cache-write tokens per test
    0
  38. 38
    Thinking MachinesDossier

    Accuracy

    Inkling Small, Accuracy: 27.6% (scale 0.0 to 100.0), 95% confidence interval 25.4 to 29.827.6%(25.4–29.8)
    Duration per test
    986.4 s
    Cost per test
    $0.40
    Input tokens per test
    5,328,516.9
    Output tokens per test
    55,978.2
    Reasoning tokens per test
    Not tested
    Cache-read tokens per test
    5,206,791.8
    Cache-write tokens per test
    0