{
  "sourceUrl": "https://github.com/Eliovp-BV/paiton-vllm-plugin/blob/8f56157c05eb6a53f6cdab00a115e47b42fed2d1/models/Qwen3.8-MXFP4-DFlash2/README.md",
  "sourceRevision": "8f56157c05eb6a53f6cdab00a115e47b42fed2d1",
  "sourceSha256": "a6c919f28ebbda238334bbdcbcdd157a4e5f610d1f9e4f9b8e1676023930667e",
  "releaseDate": "2026-09-18",
  "reviewedDate": "2026-09-19",
  "scope": "One R9700, 300 W power limit, 65K preset, BetterBench 0.6.0; 290 successful requests including warmups; thinking disabled, APC off. Not a controlled comparison with previous releases.",
  "weightedDecode": 146.9,
  "decode": [
    {
      "category": "chat",
      "tokensPerSecond": 123.7
    },
    {
      "category": "code",
      "tokensPerSecond": 169.2
    },
    {
      "category": "file_edit",
      "tokensPerSecond": 185
    },
    {
      "category": "json",
      "tokensPerSecond": 215.9
    },
    {
      "category": "math",
      "tokensPerSecond": 182.7
    },
    {
      "category": "prose",
      "tokensPerSecond": 74.9
    },
    {
      "category": "reasoning",
      "tokensPerSecond": 112.5
    },
    {
      "category": "summarization",
      "tokensPerSecond": 115
    }
  ],
  "prefill": [
    {
      "requestedDepth": 2000,
      "medianActualTokens": 1516.5,
      "tokensPerSecond": 3503.6
    },
    {
      "requestedDepth": 8000,
      "medianActualTokens": 5894.5,
      "tokensPerSecond": 3530.4
    },
    {
      "requestedDepth": 16000,
      "medianActualTokens": 11802,
      "tokensPerSecond": 3536.9
    },
    {
      "requestedDepth": 32000,
      "medianActualTokens": 23549.5,
      "tokensPerSecond": 3393
    },
    {
      "requestedDepth": 64000,
      "medianActualTokens": 47016.5,
      "tokensPerSecond": 3113
    }
  ],
  "concurrency": [
    {
      "concurrency": 1,
      "tokensPerSecond": 115
    },
    {
      "concurrency": 2,
      "tokensPerSecond": 203.2
    },
    {
      "concurrency": 4,
      "tokensPerSecond": 296.5
    },
    {
      "concurrency": 8,
      "tokensPerSecond": 400.7
    }
  ],
  "prefixProbe": {
    "configuredContext": 220000,
    "promptTokens": 215005,
    "cachedTokens": 213840,
    "coldResponseSeconds": 130,
    "repeatResponseSeconds": 1.77,
    "outputTokens": 9,
    "scope": "Single functional retrieval probe, complete response times, not TTFT or decode throughput."
  }
}
