{
  "statistic": "median",
  "excludes": [
    "unload",
    "whisper_restore",
    "queue_wait",
    "asr",
    "diarization",
    "audio_parking"
  ],
  "reference_hardware": {
    "gpu": "NVIDIA RTX 3090",
    "vram_mib": 24576,
    "cpu": "AMD Ryzen 9 5900X",
    "ram_gb": 64,
    "storage": "NVMe"
  },
  "metrics": {
    "g60_27b_q4": {
      "median_s": 8.423387409500037,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/resident-gp60-2026-10-27-09/summary/measurements.csv",
      "selection": {
        "model": "27b_q4"
      }
    },
    "g60_4b_q4": {
      "median_s": 1.505287074499961,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/resident-gp60-2026-10-27-09/summary/measurements.csv",
      "selection": {
        "model": "4b_q4"
      }
    },
    "g60_9b_q4": {
      "median_s": 2.5462516454999786,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/resident-gp60-2026-10-27-09/summary/measurements.csv",
      "selection": {
        "model": "9b_q4"
      }
    },
    "g60_9b_q6": {
      "median_s": 3.1456703310000194,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/resident-gp60-2026-10-27-09/summary/measurements.csv",
      "selection": {
        "model": "9b_q6"
      }
    },
    "g60_flash_iq2xs": {
      "median_s": 3.294117317500195,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/resident-gp60-2026-10-27-09/summary/measurements.csv",
      "selection": {
        "model": "flash_iq2xs"
      }
    },
    "min_IQ2_XS": {
      "median_s": 6.116845690001355,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/strata-iq3-2026-10-27-12/summary/measurements.csv",
      "selection": {
        "model": "IQ2_XS",
        "kind": "main"
      }
    },
    "min_IQ3_S": {
      "median_s": 8.242881229500199,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/strata-iq3-2026-10-27-12/summary/measurements.csv",
      "selection": {
        "model": "IQ3_S",
        "kind": "main"
      }
    },
    "min_IQ3_XXS": {
      "median_s": 7.160614699500002,
      "n": 60,
      "cohort": "G60",
      "includes_load": false,
      "source": "results/strata-iq3-2026-10-27-12/summary/measurements.csv",
      "selection": {
        "model": "IQ3_XXS",
        "kind": "main"
      }
    },
    "g30_27b_q4_native": {
      "median_s": 6.70904951600096,
      "n": 30,
      "cohort": "G30",
      "includes_load": false,
      "source": "results/gp30-2026-10-25-05/summary/inference_runs.csv",
      "selection": {
        "model": "27b_q4_native"
      }
    },
    "g30_4b_q4_8gb": {
      "median_s": 1.2438214715002687,
      "n": 30,
      "cohort": "G30",
      "includes_load": false,
      "source": "results/gp30-2026-10-25-05/summary/inference_runs.csv",
      "selection": {
        "model": "4b_q4_8gb"
      }
    },
    "g30_9b_q4_8gb": {
      "median_s": 2.061602458499692,
      "n": 30,
      "cohort": "G30",
      "includes_load": false,
      "source": "results/gp30-2026-10-25-05/summary/inference_runs.csv",
      "selection": {
        "model": "9b_q4_8gb"
      }
    },
    "g30_9b_q6_original": {
      "median_s": 2.48984411100173,
      "n": 30,
      "cohort": "G30",
      "includes_load": false,
      "source": "results/gp30-2026-10-25-05/summary/inference_runs.csv",
      "selection": {
        "model": "9b_q6_original"
      }
    },
    "s4": {
      "median_s": 2.2527831779989356,
      "median_load_s": 1.5194422819986357,
      "median_inference_s": 0.7332148919995234,
      "n": 5,
      "cohort": "K",
      "input_tokens": 377,
      "context": 87552,
      "includes_load": true,
      "source": "results/followup-2026-10-24-02/strict_serial/runs.json",
      "selection": {
        "model": "4b",
        "profile": "torch_limit_6gib",
        "first": false
      },
      "formula": "median(llm_load_s + llm.total_s) over paired warm runs"
    },
    "s9q4": {
      "median_s": 3.0816698229973554,
      "median_load_s": 1.8265487639982894,
      "median_inference_s": 1.2543958659989585,
      "n": 5,
      "cohort": "K",
      "input_tokens": 377,
      "context": 32768,
      "includes_load": true,
      "source": "results/q4-context-2026-10-24-03/serial/runs.json",
      "selection": {
        "context": 32768,
        "stage": "validation_short",
        "first": false
      },
      "formula": "median(llm_load_s + llm.total_s) over paired warm runs"
    },
    "s9q6": {
      "median_s": 3.447230596004374,
      "median_load_s": 2.040765032001218,
      "median_inference_s": 1.4052345390009577,
      "n": 5,
      "cohort": "K",
      "input_tokens": 377,
      "context": 87552,
      "includes_load": true,
      "source": "results/followup-2026-10-24-02/strict_serial/runs.json",
      "selection": {
        "model": "q6",
        "profile": "torch_limit_6gib",
        "first": false
      },
      "formula": "median(llm_load_s + llm.total_s) over paired warm runs"
    },
    "s27": {
      "median_s": 5.99913562599977,
      "median_load_s": 3.0642368769986206,
      "median_inference_s": 2.9341335630015237,
      "n": 5,
      "cohort": "K",
      "input_tokens": 377,
      "context": 262144,
      "includes_load": true,
      "source": "results/27b-2026-10-24-04/serial/runs.json",
      "selection": {
        "profile": "q4_native",
        "first": false
      },
      "formula": "median(llm_load_s + llm.total_s) over paired warm runs"
    }
  },
  "grid": {
    "8": {
      "16": [
        {
          "goal": "P",
          "model": "4B-Q4",
          "full_name": "Qwen3.5-4B Q4_K_M",
          "context": "87k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        }
      ],
      "32": [
        {
          "goal": "P=Q",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        }
      ],
      "48": [
        {
          "goal": "P=Q",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Test",
          "model": "Flash IQ2",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ2_XS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "O",
          "metric": null,
          "extra": "512 MiB Cache",
          "time_note": ""
        }
      ],
      "64": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "M",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q-Test",
          "model": "Flash XXS",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_XXS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "O",
          "metric": null,
          "extra": "576 MiB Cache",
          "time_note": ""
        },
        {
          "goal": "Alt.-Test",
          "model": "Flash S",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_S / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "O",
          "metric": null,
          "extra": "640 MiB Cache",
          "time_note": ""
        }
      ]
    },
    "12": {
      "16": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q6",
          "extra": "",
          "time_note": ""
        }
      ],
      "32": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q6",
          "extra": "",
          "time_note": ""
        }
      ],
      "48": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q6",
          "extra": "",
          "time_note": ""
        }
      ],
      "64": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q-Test",
          "model": "Flash XXS",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_XXS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "O",
          "metric": null,
          "extra": "576 MiB Cache",
          "time_note": ""
        },
        {
          "goal": "Alt.-Test",
          "model": "Flash S",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_S / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "O",
          "metric": null,
          "extra": "640 MiB Cache",
          "time_note": ""
        }
      ]
    },
    "16": {
      "16": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q6",
          "extra": "",
          "time_note": ""
        }
      ],
      "32": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q6",
          "extra": "",
          "time_note": ""
        }
      ],
      "48": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s9q6",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Test",
          "model": "Flash IQ2",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ2_XS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "O",
          "metric": "min_IQ2_XS",
          "extra": "512 MiB Cache",
          "time_note": ""
        }
      ],
      "64": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "M",
          "metric": "g60_9b_q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "Flash XXS",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_XXS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "M",
          "metric": "min_IQ3_XXS",
          "extra": "576 MiB Cache",
          "time_note": ""
        },
        {
          "goal": "Alt.",
          "model": "Flash S",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_S / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "M",
          "metric": "min_IQ3_S",
          "extra": "640 MiB Cache",
          "time_note": ""
        }
      ]
    },
    "20": {
      "16": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q6",
          "extra": "",
          "time_note": ""
        }
      ],
      "32": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q6",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": null,
          "extra": "",
          "time_note": "Laden + Inferenz offen"
        }
      ],
      "48": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q6",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": null,
          "extra": "",
          "time_note": "Laden + Inferenz offen"
        }
      ],
      "64": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q6",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "Flash XXS",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_XXS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "min_IQ3_XXS",
          "extra": "576 MiB Cache",
          "time_note": ""
        },
        {
          "goal": "Alt.",
          "model": "Flash S",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_S / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "min_IQ3_S",
          "extra": "640 MiB Cache",
          "time_note": ""
        },
        {
          "goal": "Alt.",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": null,
          "extra": "",
          "time_note": "Laden + Inferenz offen"
        }
      ]
    },
    "24": {
      "16": [
        {
          "goal": "P",
          "model": "9B-Q4",
          "full_name": "Qwen3.5-9B Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "A",
          "metric": "g60_9b_q4",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 3,
          "mode": "PG",
          "status": "A",
          "metric": "g30_9b_q6_original",
          "extra": "",
          "time_note": ""
        }
      ],
      "32": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 3,
          "mode": "PG",
          "status": "A",
          "metric": "g30_9b_q6_original",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "262k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s27",
          "extra": "",
          "time_note": ""
        }
      ],
      "48": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 3,
          "mode": "PG",
          "status": "A",
          "metric": "g30_9b_q6_original",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "262k",
          "slots": 1,
          "mode": "S",
          "status": "A",
          "metric": "s27",
          "extra": "",
          "time_note": ""
        }
      ],
      "64": [
        {
          "goal": "P",
          "model": "Flash IQ2",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ2_XS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "M",
          "metric": "g60_flash_iq2xs",
          "extra": "12-GiB-Prozessziel",
          "time_note": ""
        },
        {
          "goal": "Q",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "262k",
          "slots": 1,
          "mode": "S",
          "status": "M",
          "metric": "s27",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Alt.",
          "model": "Flash XXS",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_XXS / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "M",
          "metric": "min_IQ3_XXS",
          "extra": "576 MiB Cache",
          "time_note": ""
        },
        {
          "goal": "Alt.",
          "model": "Flash S",
          "full_name": "Qwen3.8-Flash-Next GSQ-RCO IQ3_S / Strata",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "M",
          "metric": "min_IQ3_S",
          "extra": "640 MiB Cache",
          "time_note": ""
        }
      ]
    },
    "32": {
      "16": [
        {
          "goal": "P=Q-Basis",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 3,
          "mode": "PG",
          "status": "A",
          "metric": "g30_9b_q6_original",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q-Test",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "H",
          "metric": null,
          "extra": "RAM-Grenze offen",
          "time_note": "RAM-Eignung offen"
        }
      ],
      "32": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 3,
          "mode": "PG",
          "status": "A",
          "metric": "g30_9b_q6_original",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q / Empfehlung",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "H",
          "metric": "g60_27b_q4",
          "extra": "",
          "time_note": ""
        }
      ],
      "48": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 3,
          "mode": "PG",
          "status": "A",
          "metric": "g30_9b_q6_original",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q / Empfehlung",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "H",
          "metric": "g60_27b_q4",
          "extra": "",
          "time_note": ""
        }
      ],
      "64": [
        {
          "goal": "P",
          "model": "9B-Q6",
          "full_name": "Qwen3.5-9B Q6_K",
          "context": "87k",
          "slots": 3,
          "mode": "PG",
          "status": "A",
          "metric": "g30_9b_q6_original",
          "extra": "",
          "time_note": ""
        },
        {
          "goal": "Q / Empfehlung",
          "model": "27B-Q4",
          "full_name": "Qwen3.8-27B UD-Q4_K_M",
          "context": "32k",
          "slots": 1,
          "mode": "PG",
          "status": "H",
          "metric": "g60_27b_q4",
          "extra": "",
          "time_note": ""
        }
      ]
    }
  }
}
