{
  "date": "2026-10-12",
  "hardware": {
    "gpu": "NVIDIA RTX 5060 Ti 16 GB",
    "cpu": "Intel Core i5-10400",
    "ramGiB": 64,
    "os": "Windows 11 Pro",
    "driver": "616.92"
  },
  "runtime": {
    "build": "llama.cpp b11429 / d81235049",
    "cuda": "13.4",
    "threads": 6,
    "batch": 512,
    "microbatch": 256,
    "flashAttention": true,
    "kvCache": "Q8",
    "repetitions": 3,
    "promptTokens": 512,
    "generatedTokens": 128
  },
  "bench": [
    {
      "quant": "Q6_K",
      "generation": 55.909717,
      "prompt": 2423.194703,
      "generationStdDev": 0.072438
    },
    {
      "quant": "Q8_0",
      "generation": 44.648246,
      "prompt": 2925.454454,
      "generationStdDev": 0.076001
    }
  ],
  "contexts": [
    {
      "context": 4096,
      "promptTokens": 3513,
      "q6": {
        "generation": 53.29,
        "prompt": 2343.61,
        "firstTokenSeconds": 1.57,
        "gpuMiB": 13747
      },
      "q8": {
        "generation": 43.44,
        "prompt": 2835.8,
        "firstTokenSeconds": 1.33,
        "gpuMiB": 15863
      }
    },
    {
      "context": 8192,
      "promptTokens": 7609,
      "q6": {
        "generation": 52.48,
        "prompt": 2329.12,
        "firstTokenSeconds": 3.32,
        "gpuMiB": 13693
      },
      "q8": {
        "generation": 42.44,
        "prompt": 2754.33,
        "firstTokenSeconds": 2.82,
        "gpuMiB": 15765
      }
    },
    {
      "context": 16384,
      "promptTokens": 15801,
      "q6": {
        "generation": 50.93,
        "prompt": 2287.63,
        "firstTokenSeconds": 6.97,
        "gpuMiB": 13864
      },
      "q8": {
        "generation": 41.19,
        "prompt": 2723.54,
        "firstTokenSeconds": 5.87,
        "gpuMiB": 15771
      }
    }
  ],
  "notes": [
    "Q8_0 results are historical; Q6_K is the current configuration.",
    "GPU figures are total-device snapshots including other applications, not peak measurements.",
    "Context tests are one sample per context, not prolonged stability tests.",
    "Quality equivalence has not been tested."
  ]
}