{
  "title": "Every number I had said Ornith",
  "footnote": "Median of three with a warmup call, thinking off, streamed. GLM-5.3-Flash's middle reading was at 21K, and it has no 151K reading.",
  "kind": "matrix",
  "options": {
    "mode": "value",
    "max": 172.7,
    "columns": [
      {
        "key": "a",
        "label": "2.4K"
      },
      {
        "key": "b",
        "label": "38K"
      },
      {
        "key": "c",
        "label": "151K"
      }
    ],
    "decimals": 0
  },
  "rows": [
    {
      "label": "Ornith-1.5-35B-A3B",
      "note": "int4, vLLM, PRO 6000",
      "a": 143.8,
      "b": 172.7,
      "c": 159.1
    },
    {
      "label": "Ornith-1.5",
      "note": "int4, vLLM, 2x Strix",
      "a": 103,
      "b": 136,
      "c": 102
    },
    {
      "label": "Laguna-S-2.1",
      "note": "Q4_K_XL, llama.cpp",
      "a": 77.5,
      "b": 99.6,
      "c": 72.1
    },
    {
      "label": "Qwen3.8-Flash-Next",
      "note": "Q3_K_XL, llama.cpp",
      "a": 70,
      "b": 54,
      "c": 28
    },
    {
      "label": "Qwen3.8-27B dense",
      "note": "FP8, vLLM",
      "a": 46.1,
      "b": 48.2,
      "c": 40.2
    },
    {
      "label": "GLM-5.3-Flash",
      "note": "IQ2_XXS, llama.cpp fork",
      "a": 32.7,
      "b": 19.7
    }
  ],
  "series": [
    {
      "key": "a",
      "label": "2.4K"
    },
    {
      "key": "b",
      "label": "38K"
    },
    {
      "key": "c",
      "label": "151K"
    }
  ],
  "subtitle": "Single-stream decode in tokens per second, by prompt size. Brighter is faster.",
  "xLabel": "Model",
  "__hash__": "uZO5XPbmj66qrgRvEWjxq8SzarMD9_GbYBnrvFv5X28"
}