{
  "schema": "l2s1.rtx3060.public-aggregates.v1",
  "date": "2026-09-23",
  "exportDate": "2026-09-26",
  "dataset": "JevBench v1.1 public 231",
  "gpu": "NVIDIA GeForce RTX 3060",
  "vramMiB": 12288,
  "driver": "595.71.05",
  "context": 8192,
  "models": 20,
  "decisions": 4620,
  "itemsPerModel": 231,
  "tierCounts": {
    "easy": 48,
    "original": 72,
    "hard": 111
  },
  "policy": {
    "minTopProbability": 0.8,
    "minCandidateMass": 0.05
  },
  "latencyScope": "Rust serial decide_batch, batch size 1; model load and one warmup excluded",
  "fileSizeScope": "GGUF model file bytes; frozen download metadata or SHA-256-verified local file stat",
  "gpuMemoryScope": "Maximum sampled whole-device memory used, 200 ms sampling",
  "jevbenchRevision": "f79a1cab94ab9a5879383b7ef9ee1805b9dc2d84",
  "evaluatorSha256": "312ceabd7f82ed769d97eb5528a726de6ad00dd29ed9585791d5a466280e45a8",
  "totals": {
    "planned": 4620,
    "attempted": 4620,
    "scorable": 4620,
    "valid": 4620,
    "rawCorrect": 2924,
    "accepted": 3122,
    "acceptedCorrect": 2279,
    "acceptedWrong": 843,
    "abstained": 1498,
    "errors": 0
  },
  "rows": [
    {
      "fileSizeBytes": 4482403488,
      "id": "Qwen3.5-4B-Q8_0",
      "model": "Qwen3.5 4B Q8_0",
      "modelFile": "Qwen3.5-4B-Q8_0.gguf",
      "rawAccuracy": 79.65367965367966,
      "coverage": 59.74025974025974,
      "acceptedAccuracy": 93.47826086956522,
      "correctAll": 55.84415584415584,
      "p50Ms": 86.875174,
      "p95Ms": 1137.298412,
      "peakGpuMiB": 5039.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 184,
        "accepted": 138,
        "acceptedCorrect": 129,
        "acceptedWrong": 9,
        "abstained": 93,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 5680522464,
      "id": "Qwen3.5-9B-Q4_K_M",
      "model": "Qwen3.5 9B Q4_K_M",
      "modelFile": "Qwen3.5-9B-Q4_K_M.gguf",
      "rawAccuracy": 77.92207792207793,
      "coverage": 72.2943722943723,
      "acceptedAccuracy": 92.21556886227546,
      "correctAll": 66.66666666666666,
      "p50Ms": 130.55342000000002,
      "p95Ms": 1697.1431205,
      "peakGpuMiB": 5637.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 180,
        "accepted": 167,
        "acceptedCorrect": 154,
        "acceptedWrong": 13,
        "abstained": 64,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 9527502048,
      "id": "Qwen3.5-9B-Q8_0",
      "model": "Qwen3.5 9B Q8_0",
      "modelFile": "Qwen3.5-9B-Q8_0.gguf",
      "rawAccuracy": 77.92207792207793,
      "coverage": 71.42857142857143,
      "acceptedAccuracy": 91.51515151515152,
      "correctAll": 65.36796536796537,
      "p50Ms": 124.679976,
      "p95Ms": 1613.7917155,
      "peakGpuMiB": 8817.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 180,
        "accepted": 165,
        "acceptedCorrect": 151,
        "acceptedWrong": 14,
        "abstained": 66,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 4977171584,
      "id": "gemma-4-E4B-it-Q4_K_M",
      "model": "Gemma 4 E4B it Q4_K_M",
      "modelFile": "gemma-4-E4B-it-Q4_K_M.gguf",
      "rawAccuracy": 77.48917748917748,
      "coverage": 83.98268398268398,
      "acceptedAccuracy": 84.5360824742268,
      "correctAll": 70.995670995671,
      "p50Ms": 86.620102,
      "p95Ms": 1186.4660155,
      "peakGpuMiB": 3989.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 179,
        "accepted": 194,
        "acceptedCorrect": 164,
        "acceptedWrong": 30,
        "abstained": 37,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 8192953472,
      "id": "gemma-4-E4B-it-Q8_0",
      "model": "Gemma 4 E4B it Q8_0",
      "modelFile": "gemma-4-E4B-it-Q8_0.gguf",
      "rawAccuracy": 76.62337662337663,
      "coverage": 83.98268398268398,
      "acceptedAccuracy": 83.50515463917526,
      "correctAll": 70.12987012987013,
      "p50Ms": 85.03775399999999,
      "p95Ms": 1146.213878,
      "peakGpuMiB": 6047.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 177,
        "accepted": 194,
        "acceptedCorrect": 162,
        "acceptedWrong": 32,
        "abstained": 37,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 2740937888,
      "id": "Qwen3.5-4B-Q4_K_M",
      "model": "Qwen3.5 4B Q4_K_M",
      "modelFile": "Qwen3.5-4B-Q4_K_M.gguf",
      "rawAccuracy": 76.19047619047619,
      "coverage": 60.60606060606061,
      "acceptedAccuracy": 92.85714285714286,
      "correctAll": 56.27705627705628,
      "p50Ms": 88.46956,
      "p95Ms": 1175.1838305,
      "peakGpuMiB": 3381.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 176,
        "accepted": 140,
        "acceptedCorrect": 130,
        "acceptedWrong": 10,
        "abstained": 91,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 7266070528,
      "id": "Qwen3.8-27B-UD-IQ2_XXS",
      "model": "Qwen3.8 27B UD-IQ2_XXS",
      "modelFile": "Qwen3.8-27B-UD-IQ2_XXS.gguf",
      "rawAccuracy": 73.16017316017316,
      "coverage": 63.63636363636363,
      "acceptedAccuracy": 87.75510204081633,
      "correctAll": 55.84415584415584,
      "p50Ms": 414.60704999999996,
      "p95Ms": 5460.217004,
      "peakGpuMiB": 7825.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 169,
        "accepted": 147,
        "acceptedCorrect": 129,
        "acceptedWrong": 18,
        "abstained": 84,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 8709519168,
      "id": "Qwen3-8B-Q8_0",
      "model": "Qwen3 8B Q8_0",
      "modelFile": "Qwen3-8B-Q8_0.gguf",
      "rawAccuracy": 71.42857142857143,
      "coverage": 93.5064935064935,
      "acceptedAccuracy": 74.07407407407408,
      "correctAll": 69.26406926406926,
      "p50Ms": 121.450278,
      "p95Ms": 1817.940254,
      "peakGpuMiB": 9347.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 165,
        "accepted": 216,
        "acceptedCorrect": 160,
        "acceptedWrong": 56,
        "abstained": 15,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 4967497152,
      "id": "gemma-4-E2B-it-Q8_0",
      "model": "Gemma 4 E2B it Q8_0",
      "modelFile": "gemma-4-E2B-it-Q8_0.gguf",
      "rawAccuracy": 67.96536796536796,
      "coverage": 90.47619047619048,
      "acceptedAccuracy": 72.24880382775119,
      "correctAll": 65.36796536796537,
      "p50Ms": 46.787102,
      "p95Ms": 701.11393,
      "peakGpuMiB": 3025.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 157,
        "accepted": 209,
        "acceptedCorrect": 151,
        "acceptedWrong": 58,
        "abstained": 22,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 4280405792,
      "id": "Qwen3-4B-Q8_0",
      "model": "Qwen3 4B Q8_0",
      "modelFile": "Qwen3-4B-Q8_0.gguf",
      "rawAccuracy": 65.80086580086581,
      "coverage": 87.44588744588745,
      "acceptedAccuracy": 68.8118811881188,
      "correctAll": 60.17316017316018,
      "p50Ms": 85.26035999999999,
      "p95Ms": 1349.2774075000002,
      "peakGpuMiB": 5751.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 152,
        "accepted": 202,
        "acceptedCorrect": 139,
        "acceptedWrong": 63,
        "abstained": 29,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 5198386720,
      "id": "Ministral-3-8B-Instruct-2512-Q4_K_M",
      "model": "Ministral-3 8B Instruct 2512 Q4_K_M",
      "modelFile": "Ministral-3-8B-Instruct-2512-Q4_K_M.gguf",
      "rawAccuracy": 65.36796536796537,
      "coverage": 59.74025974025974,
      "acceptedAccuracy": 83.33333333333334,
      "correctAll": 49.78354978354979,
      "p50Ms": 441.42554,
      "p95Ms": 2399.6961725,
      "peakGpuMiB": 6281.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 151,
        "accepted": 138,
        "acceptedCorrect": 115,
        "acceptedWrong": 23,
        "abstained": 93,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 11624759488,
      "id": "gpt-oss-20b-Q4_K_M-no-cuda-graphs",
      "model": "gpt-oss 20B Q4_K_M",
      "modelFile": "gpt-oss-20b-Q4_K_M.gguf",
      "rawAccuracy": 64.93506493506493,
      "coverage": 64.5021645021645,
      "acceptedAccuracy": 79.19463087248322,
      "correctAll": 51.082251082251084,
      "p50Ms": 181.931967,
      "p95Ms": 2262.284453,
      "peakGpuMiB": 11901.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 150,
        "accepted": 149,
        "acceptedCorrect": 118,
        "acceptedWrong": 31,
        "abstained": 82,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {
          "GGML_CUDA_DISABLE_GRAPHS": "1"
        }
      }
    },
    {
      "fileSizeBytes": 2012012800,
      "id": "Qwen3.5-2B-Q8_0",
      "model": "Qwen3.5 2B Q8_0",
      "modelFile": "Qwen3.5-2B-Q8_0.gguf",
      "rawAccuracy": 62.33766233766234,
      "coverage": 42.42424242424242,
      "acceptedAccuracy": 84.6938775510204,
      "correctAll": 35.93073593073593,
      "p50Ms": 40.992193,
      "p95Ms": 513.032822,
      "peakGpuMiB": 2465.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 144,
        "accepted": 98,
        "acceptedCorrect": 83,
        "acceptedWrong": 15,
        "abstained": 133,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 4130402176,
      "id": "gemma-3-4b-it-Q8_0",
      "model": "Gemma 3 4B it Q8_0",
      "modelFile": "gemma-3-4b-it-Q8_0.gguf",
      "rawAccuracy": 59.74025974025974,
      "coverage": 96.1038961038961,
      "acceptedAccuracy": 60.36036036036037,
      "correctAll": 58.00865800865801,
      "p50Ms": 74.710081,
      "p95Ms": 879.920001,
      "peakGpuMiB": 5477.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 138,
        "accepted": 222,
        "acceptedCorrect": 134,
        "acceptedWrong": 88,
        "abstained": 9,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 4084611040,
      "id": "Phi-4-mini-instruct.Q8_0",
      "model": "Phi-4-mini Instruct Q8_0",
      "modelFile": "Phi-4-mini-instruct.Q8_0.gguf",
      "rawAccuracy": 56.709956709956714,
      "coverage": 50.21645021645021,
      "acceptedAccuracy": 74.13793103448276,
      "correctAll": 37.22943722943723,
      "p50Ms": 70.499215,
      "p95Ms": 971.9538835,
      "peakGpuMiB": 5337.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 131,
        "accepted": 116,
        "acceptedCorrect": 86,
        "acceptedWrong": 30,
        "abstained": 115,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 811843840,
      "id": "Qwen3.5-0.8B-Q8_0",
      "model": "Qwen3.5 0.8B Q8_0",
      "modelFile": "Qwen3.5-0.8B-Q8_0.gguf",
      "rawAccuracy": 51.94805194805194,
      "coverage": 20.77922077922078,
      "acceptedAccuracy": 66.66666666666666,
      "correctAll": 13.852813852813853,
      "p50Ms": 27.927573000000002,
      "p95Ms": 350.530776,
      "peakGpuMiB": 1317.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 120,
        "accepted": 48,
        "acceptedCorrect": 32,
        "acceptedWrong": 16,
        "abstained": 183,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 3275575840,
      "id": "SmolLM3-3B-Q8_0",
      "model": "SmolLM3 3B Q8_0",
      "modelFile": "SmolLM3-3B-Q8_0.gguf",
      "rawAccuracy": 46.320346320346324,
      "coverage": 45.02164502164502,
      "acceptedAccuracy": 60.57692307692307,
      "correctAll": 27.27272727272727,
      "p50Ms": 71.471313,
      "p95Ms": 884.7550485,
      "peakGpuMiB": 4047.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 107,
        "accepted": 104,
        "acceptedCorrect": 63,
        "acceptedWrong": 41,
        "abstained": 127,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 3421898816,
      "id": "Llama-3.2-3B-Instruct-Q8_0",
      "model": "Llama-3.2 3B Instruct Q8_0",
      "modelFile": "Llama-3.2-3B-Instruct-Q8_0.gguf",
      "rawAccuracy": 43.722943722943725,
      "coverage": 36.36363636363637,
      "acceptedAccuracy": 45.23809523809524,
      "correctAll": 16.450216450216452,
      "p50Ms": 68.183239,
      "p95Ms": 884.752845,
      "peakGpuMiB": 4585.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 101,
        "accepted": 84,
        "acceptedCorrect": 38,
        "acceptedWrong": 46,
        "abstained": 147,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 1069306368,
      "id": "gemma-3-1b-it-Q8_0",
      "model": "Gemma 3 1B it Q8_0",
      "modelFile": "gemma-3-1b-it-Q8_0.gguf",
      "rawAccuracy": 38.961038961038966,
      "coverage": 87.01298701298701,
      "acceptedAccuracy": 39.800995024875625,
      "correctAll": 34.63203463203463,
      "p50Ms": 25.056749,
      "p95Ms": 309.2379475,
      "peakGpuMiB": 1663.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 90,
        "accepted": 201,
        "acceptedCorrect": 80,
        "acceptedWrong": 121,
        "abstained": 30,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    },
    {
      "fileSizeBytes": 639446688,
      "id": "Qwen3-0.6B-Q8_0",
      "model": "Qwen3 0.6B Q8_0",
      "modelFile": "Qwen3-0.6B-Q8_0.gguf",
      "rawAccuracy": 31.601731601731604,
      "coverage": 82.25108225108225,
      "acceptedAccuracy": 32.10526315789474,
      "correctAll": 26.406926406926406,
      "p50Ms": 34.014483,
      "p95Ms": 445.11573999999996,
      "peakGpuMiB": 1857.0,
      "valid": 231,
      "errors": 0,
      "counts": {
        "planned": 231,
        "attempted": 231,
        "scorable": 231,
        "valid": 231,
        "rawCorrect": 73,
        "accepted": 190,
        "acceptedCorrect": 61,
        "acceptedWrong": 129,
        "abstained": 41,
        "errors": 0
      },
      "config": {
        "context": 8192,
        "batch": 256,
        "microbatch": 256,
        "threads": 4,
        "flashAttention": "off",
        "executionMode": "fresh",
        "promptLayout": "legacy",
        "runtimeEnvironmentOverrides": {}
      }
    }
  ]
}
