Acropolis

LMArena Agent

Arena agent leaderboard: score with confidence interval, cost and output tokens per task, observation and session counts. One row per model and category.

Upstream source
AgenticCoding
Independent. Run by an operator independent of the model vendors.LMArena by ArenaVersion 1Data as of 17 Nov 2026

No filters. Showing 64 of 64 rows.

  1. 1
    AnthropicDossier

    max

    Agent score

    Fable 5.2, Agent score: 0.3 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.2 to 0.30.3(0.2–0.3)
    Cost per task (median)
    $9.02
    Output tokens per task (median)
    126,742.5
  2. 2
    AnthropicDossier

    max

    Agent score

    Opus 5.5, Agent score: 0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.2 to 0.20.2(0.2–0.2)
    Cost per task (median)
    $5.65
    Output tokens per task (median)
    107,510.5
  3. 3
    GoogleDossier

    high

    Agent score

    Gemini 4 Carbon, Agent score: 0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.2 to 0.20.2(0.2–0.2)
    Cost per task (median)
    $1.35
    Output tokens per task (median)
    44,991
  4. 4
    AnthropicDossier

    max

    Agent score

    Sonnet 5.5, Agent score: 0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.20.2(0.1–0.2)
    Cost per task (median)
    $3.46
    Output tokens per task (median)
    150,642
  5. 5
    GoogleDossier

    high

    Agent score

    Gemini 4 Hydrogen, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.20.1(0.1–0.2)
    Cost per task (median)
    $1.53
    Output tokens per task (median)
    64,700
  6. 6
    AnthropicDossier

    max

    Agent score

    Fable 5.1, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.20.1(0.1–0.2)
    Cost per task (median)
    $7.83
    Output tokens per task (median)
    59,147.5
  7. 7
    OpenAIDossier

    max

    Agent score

    GPT-6 Astra, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.20.1(0.1–0.2)
    Cost per task (median)
    $5.28
    Output tokens per task (median)
    32,278
  8. 8
    AnthropicDossier

    high

    Agent score

    Opus 5.5, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    $2.12
    Output tokens per task (median)
    31,743
  9. 9
    OpenAIDossier

    max

    Agent score

    GPT-6.1 Sol, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    $0.99
    Output tokens per task (median)
    33,771.5
  10. 10
    OpenAIDossier

    max

    Agent score

    GPT-6.1 Terra, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    $0.87
    Output tokens per task (median)
    45,319.5
  11. 11
    GoogleDossier

    high

    Agent score

    Gemini 4 Argon, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    $0.70
    Output tokens per task (median)
    22,853.5
  12. 12
    OpenAIDossier

    medium

    Agent score

    GPT-6.1 Sol, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    $0.37
    Output tokens per task (median)
    8,525
  13. 13

    max

    Agent score

    Muse Wonder Full 1.0, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    Not tested
    Output tokens per task (median)
    29,708.3
  14. 14
    AnthropicDossier

    high

    Agent score

    Sonnet 5.5, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    $0.88
    Output tokens per task (median)
    22,997
  15. 15
    OpenAIDossier

    max

    Agent score

    GPT-6 Sol, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.10.1(0.0–0.1)
    Cost per task (median)
    $1.85
    Output tokens per task (median)
    39,154.5
  16. 16
    AnthropicDossier

    high

    Agent score

    Fable 5, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.1 to 0.10.1(0.1–0.1)
    Cost per task (median)
    $3.33
    Output tokens per task (median)
    22,867.5
  17. 17
    AnthropicDossier

    max

    Agent score

    Opus 5, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.10.1(0.0–0.1)
    Cost per task (median)
    $4.50
    Output tokens per task (median)
    61,124.3
  18. 18
    TencentDossier

    Agent score

    Hy4, Agent score: 0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.10.1(0.0–0.1)
    Cost per task (median)
    $0.57
    Output tokens per task (median)
    59,209
  19. 19
    OpenAIDossier

    medium

    Agent score

    GPT-6.1 Terra, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.10(0.0–0.1)
    Cost per task (median)
    $0.16
    Output tokens per task (median)
    6,937.5
  20. 20
    Moonshot AIDossier

    max

    Agent score

    Kimi K3.1 Pro, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.10(0.0–0.1)
    Cost per task (median)
    $1.16
    Output tokens per task (median)
    32,837
  21. 21
    OpenAIDossier

    xhigh

    Agent score

    GPT-5.6 Sol, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $1.59
    Output tokens per task (median)
    22,826
  22. 22
    Thinking MachinesDossier

    Agent score

    Inkling 2 Small, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $0.26
    Output tokens per task (median)
    49,485
  23. 23

    xhigh

    Agent score

    Grok 4.7, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $3.20
    Output tokens per task (median)
    77,876
  24. 24

    max

    Agent score

    Muse Ember, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $0.020
    Output tokens per task (median)
    10,083
  25. 25
    XiaomiDossier

    Agent score

    MiMo-V2.6-Pro, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $0.16
    Output tokens per task (median)
    47,607.5
  26. 26
    GoogleDossier

    high

    Agent score

    Gemini 3.8 Flash, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $0.76
    Output tokens per task (median)
    57,563
  27. 27

    max

    Agent score

    Muse Spark 1.3, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $0.47
    Output tokens per task (median)
    28,118
  28. 28
    AnthropicDossier

    high

    Agent score

    Sonnet 5, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.00(-0.0–0.0)
    Cost per task (median)
    $0.87
    Output tokens per task (median)
    23,366
  29. 29
    DeepSeekDossier

    max

    Agent score

    DeepSeek V4.1 Flash 1017, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $0.14
    Output tokens per task (median)
    88,678.5
  30. 30

    max

    Agent score

    Muse Wonder 1.0, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.00(-0.0–0.0)
    Cost per task (median)
    $1.16
    Output tokens per task (median)
    22,992
  31. 31
    OpenAIDossier

    max

    Agent score

    GPT-6 Luna, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval 0.0 to 0.00(0.0–0.0)
    Cost per task (median)
    $0.083
    Output tokens per task (median)
    41,083
  32. 32

    max

    Agent score

    GLM-5.3, Agent score: 0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.00(-0.0–0.0)
    Cost per task (median)
    $0.86
    Output tokens per task (median)
    61,631.5
  33. 33
    Moonshot AIDossier

    max

    Agent score

    Kimi K3, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.0-0(-0.0–0.0)
    Cost per task (median)
    $0.89
    Output tokens per task (median)
    25,449.5
  34. 34
    TencentDossier

    Agent score

    Hy4 Preview, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.0-0(-0.0–0.0)
    Cost per task (median)
    $0.43
    Output tokens per task (median)
    49,617
  35. 35
    AnthropicDossier

    high

    Agent score

    Haiku 5.5, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.0-0(-0.0–0.0)
    Cost per task (median)
    $0.085
    Output tokens per task (median)
    26,659
  36. 36

    max

    Agent score

    GLM-5.2, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.0-0(-0.0–0.0)
    Cost per task (median)
    $0.86
    Output tokens per task (median)
    52,309
  37. 37
    XiaomiDossier

    Agent score

    MiMo-V2.6-Flash, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.0-0(-0.0–0.0)
    Cost per task (median)
    $0.075
    Output tokens per task (median)
    44,545
  38. 38
    AlibabaDossier

    Agent score

    Qwen3.8-Max, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to -0.0-0(-0.0–-0.0)
    Cost per task (median)
    $1.24
    Output tokens per task (median)
    50,635
  39. 39

    xhigh

    Agent score

    Grok 4.6, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to 0.0-0(-0.0–0.0)
    Cost per task (median)
    $1.33
    Output tokens per task (median)
    33,480
  40. 40
    OpenAIDossier

    xhigh

    Agent score

    GPT-5.6 Terra, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to -0.0-0(-0.0–-0.0)
    Cost per task (median)
    $0.63
    Output tokens per task (median)
    17,989
  41. 41
    OpenAIDossier

    medium

    Agent score

    GPT-6 Luna, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to -0.0-0(-0.0–-0.0)
    Cost per task (median)
    $0.025
    Output tokens per task (median)
    9,507
  42. 42
    GoogleDossier

    high

    Agent score

    Gemini 3.7 Flash, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to -0.0-0(-0.0–-0.0)
    Cost per task (median)
    $0.49
    Output tokens per task (median)
    42,886
  43. 43
    MiniMaxDossier

    Agent score

    MiniMax M3.1 Flash, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to -0.0-0(-0.0–-0.0)
    Cost per task (median)
    $0.21
    Output tokens per task (median)
    69,867
  44. 44

    Agent score

    GLM-5.3 Flash, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.0 to -0.0-0(-0.0–-0.0)
    Cost per task (median)
    $0.14
    Output tokens per task (median)
    46,332
  45. 45
    StepFunDossier

    Agent score

    Step 5 Preview, Agent score: -0 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.0-0(-0.1–-0.0)
    Cost per task (median)
    $0.70
    Output tokens per task (median)
    47,499
  46. 46
    AlibabaDossier

    Agent score

    Qwen3.8 Flash Next, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.0-0.1(-0.1–-0.0)
    Cost per task (median)
    $0.23
    Output tokens per task (median)
    63,527
  47. 47
    AlibabaDossier

    Agent score

    Qwen3.8-27B, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.0-0.1(-0.1–-0.0)
    Cost per task (median)
    $0.39
    Output tokens per task (median)
    53,374
  48. 48
    GoogleDossier

    Agent score

    Gemini 3.1 Pro Preview, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.0-0.1(-0.1–-0.0)
    Cost per task (median)
    $0.59
    Output tokens per task (median)
    13,424.5
  49. 49
    MistralDossier

    Agent score

    Mistral Large 4, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.0-0.1(-0.1–-0.0)
    Cost per task (median)
    $0.32
    Output tokens per task (median)
    26,158.8
  50. 50
    OpenAIDossier

    low

    Agent score

    GPT-6 Luna, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    $0.010
    Output tokens per task (median)
    3,746
  51. 51
    ReflectionDossier

    Agent score

    Beam 1.0, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.0-0.1(-0.1–-0.0)
    Cost per task (median)
    Not tested
    Output tokens per task (median)
    15,788
  52. 52
    AlibabaDossier

    Agent score

    Qwen3.7 Max, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    $0.28
    Output tokens per task (median)
    15,367
  53. 53
    TencentDossier

    Agent score

    Hy3, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    $0.083
    Output tokens per task (median)
    31,696
  54. 54

    Agent score

    Muse Spark 1.1, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    Not tested
    Output tokens per task (median)
    Not tested
  55. 55
    AlibabaDossier

    Agent score

    Qwen3.7 Plus, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    $0.17
    Output tokens per task (median)
    17,366
  56. 56

    xhigh

    Agent score

    Muse Spark 1.2, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    $0.41
    Output tokens per task (median)
    25,558
  57. 57
    MiniMaxDossier

    Agent score

    MiniMax M3, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    $0.38
    Output tokens per task (median)
    26,416
  58. 58
    SynlanDossier

    Agent score

    Spark-X2.6, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.1 to -0.1-0.1(-0.1–-0.1)
    Cost per task (median)
    $0.083
    Output tokens per task (median)
    33,500
  59. 59
    MistralDossier

    Agent score

    Mistral Medium 3.5, Agent score: -0.1 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.2 to -0.1-0.1(-0.2–-0.1)
    Cost per task (median)
    $0.30
    Output tokens per task (median)
    12,776
  60. 60
    NVIDIADossier

    Agent score

    Nemotron 3 Ultra 550B A55B, Agent score: -0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.2 to -0.1-0.2(-0.2–-0.1)
    Cost per task (median)
    $0.13
    Output tokens per task (median)
    8,222
  61. 61
    NVIDIADossier

    Agent score

    Nemotron 3.5 Super VL 120B A12B (NVFP4), Agent score: -0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.2 to -0.2-0.2(-0.2–-0.2)
    Cost per task (median)
    Not tested
    Output tokens per task (median)
    14,558.5
  62. 62
    Thinking MachinesDossier

    Agent score

    Inkling, Agent score: -0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.2 to -0.2-0.2(-0.2–-0.2)
    Cost per task (median)
    $0.30
    Output tokens per task (median)
    14,214.5
  63. 63
    Thinking MachinesDossier

    Agent score

    Inkling Small, Agent score: -0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.2 to -0.2-0.2(-0.2–-0.2)
    Cost per task (median)
    $0.16
    Output tokens per task (median)
    20,902
  64. 64
    UpstageDossier

    Agent score

    Solar Pro 4, Agent score: -0.2 (scale -1000000.0 to 1000000.0), 95% confidence interval -0.3 to -0.2-0.2(-0.3–-0.2)
    Cost per task (median)
    $0.041
    Output tokens per task (median)
    13,681
Unmatched upstream rows excluded from this board (3)

These upstream rows could not be mapped cleanly to the canonical model registry or reasoning effort enum.

Raw nameParsed modelReasonDetail
Qwen4-Max (xHigh)Qwen4-Maxno-aliasscore
Qwen4-Max (xHigh)Qwen4-Maxno-aliascost-per-task
Qwen4-Max (xHigh)Qwen4-Maxno-aliasoutput-tokens-per-task