{
  "schema": "anvil-serving.benchmark-graph-manifest/v1",
  "title": "GLM-5.3-Flash: 60-request endurance comparison",
  "subtitle": "Warm direct, TP2/C1, 4K target; n=60 per platform; variable short outputs and shared prefixes",
  "output": "endurance-matrix.svg",
  "data_output": "endurance-graph-data.json",
  "charts": [
    {
      "title": "Decode median: higher is faster",
      "metric": "metrics.decode_tok_s_p50",
      "x_label": "4K nominal prompt target",
      "y_label": "tokens / second",
      "series": [
        {
          "label": "Windows / WSL2",
          "points": [
            {
              "x": "60 requests",
              "artifact": "windows-endurance-4k-r60.json"
            }
          ]
        },
        {
          "label": "Native Linux",
          "points": [
            {
              "x": "60 requests",
              "artifact": "endurance-4k-r60.json"
            }
          ]
        }
      ]
    },
    {
      "title": "First visible token median: lower is faster",
      "metric": "metrics.ttft_p50_ms",
      "x_label": "4K nominal prompt target",
      "y_label": "milliseconds",
      "series": [
        {
          "label": "Windows / WSL2",
          "points": [
            {
              "x": "60 requests",
              "artifact": "windows-endurance-4k-r60.json"
            }
          ]
        },
        {
          "label": "Native Linux",
          "points": [
            {
              "x": "60 requests",
              "artifact": "endurance-4k-r60.json"
            }
          ]
        }
      ]
    }
  ]
}