{
  "schema": "amdai-bench",
  "version": 1,
  "as_of": "2026-10-13",
  "label": "Measured on our rig, 2026-10",
  "source": "local desktop official run cards (amdai-bench/run@1)",
  "results": [
    {
      "id": "local-desktop-qwen36-ud-q5-k-xl",
      "trust": "verified",
      "hardware": {
        "host": "local-desktop",
        "gpu": "Radeon AI PRO R9700",
        "gpu_id": "r9700",
        "gfx": "gfx1201",
        "gpu_count": 1,
        "gpus_used": 1,
        "vram_gb_each": 32,
        "ram_gb": 256,
        "fingerprint": "local-desktop|r9700|gfx1201|32|256|1"
      },
      "pins": {
        "model": "Qwen3.6-35B-A3B",
        "official": "unsloth/Qwen3.6-35B-A3B",
        "gguf": "unsloth/Qwen3.6-35B-A3B-GGUF",
        "quant": "UD-Q5_K_XL",
        "backend": "llama.cpp",
        "runtime": "Vulkan",
        "kv": "f16",
        "offload": "llama.cpp --fit-target 1024 MiB placed tensors; experts_to_ram inferred from model size vs peak VRAM",
        "mtp": false,
        "threads": null,
        "recipe": null
      },
      "contexts": [
        {
          "tokens": 4096,
          "label": "short",
          "prefill_tok_s": 2761.6,
          "decode_tok_s": 102.39,
          "vram_gb": 25.97
        },
        {
          "tokens": 32768,
          "label": "32K",
          "prefill_tok_s": 1937.9,
          "decode_tok_s": 90.52,
          "vram_gb": 26.7
        },
        {
          "tokens": 131072,
          "label": "128K",
          "prefill_tok_s": 942.8,
          "decode_tok_s": 66.43,
          "vram_gb": 28.82
        },
        {
          "tokens": 262144,
          "label": "256K",
          "prefill_tok_s": 519.2,
          "decode_tok_s": 44.4,
          "vram_gb": 29.34
        }
      ]
    },
    {
      "id": "local-desktop-qwen36-q6-k",
      "trust": "verified",
      "hardware": {
        "host": "local-desktop",
        "gpu": "Radeon AI PRO R9700",
        "gpu_id": "r9700",
        "gfx": "gfx1201",
        "gpu_count": 1,
        "gpus_used": 1,
        "vram_gb_each": 32,
        "ram_gb": 256,
        "fingerprint": "local-desktop|r9700|gfx1201|32|256|1"
      },
      "pins": {
        "model": "Qwen3.6-35B-A3B",
        "official": "bartowski/Qwen_Qwen3.6-35B-A3B",
        "gguf": "bartowski/Qwen_Qwen3.6-35B-A3B-GGUF",
        "quant": "Q6_K",
        "backend": "llama.cpp",
        "runtime": "Vulkan",
        "kv": "f16",
        "offload": "llama.cpp --fit-target 1024 MiB placed tensors; experts_to_ram inferred from model size vs peak VRAM",
        "mtp": false,
        "threads": null,
        "recipe": null
      },
      "contexts": [
        {
          "tokens": 4096,
          "label": "short",
          "prefill_tok_s": 2621.7,
          "decode_tok_s": 82.62,
          "vram_gb": 29.32
        },
        {
          "tokens": 4096,
          "label": "short",
          "prefill_tok_s": 2623.6,
          "decode_tok_s": 103.23,
          "vram_gb": 29.26
        },
        {
          "tokens": 32768,
          "label": "32K",
          "prefill_tok_s": 1683,
          "decode_tok_s": 71.19,
          "vram_gb": 29.17
        },
        {
          "tokens": 131072,
          "label": "128K",
          "prefill_tok_s": 800.3,
          "decode_tok_s": 54.51,
          "vram_gb": 29.25
        },
        {
          "tokens": 262144,
          "label": "256K",
          "prefill_tok_s": 471.6,
          "decode_tok_s": 37.78,
          "vram_gb": 29.3
        }
      ]
    },
    {
      "id": "local-desktop-qwen36-q5-k-m",
      "trust": "verified",
      "hardware": {
        "host": "local-desktop",
        "gpu": "Radeon AI PRO R9700",
        "gpu_id": "r9700",
        "gfx": "gfx1201",
        "gpu_count": 1,
        "gpus_used": 1,
        "vram_gb_each": 32,
        "ram_gb": 256,
        "fingerprint": "local-desktop|r9700|gfx1201|32|256|1"
      },
      "pins": {
        "model": "Qwen3.6-35B-A3B",
        "official": "bartowski/Qwen_Qwen3.6-35B-A3B",
        "gguf": "bartowski/Qwen_Qwen3.6-35B-A3B-GGUF",
        "quant": "Q5_K_M",
        "backend": "llama.cpp",
        "runtime": "Vulkan",
        "kv": "f16",
        "offload": "llama.cpp --fit-target 1024 MiB placed tensors; experts_to_ram inferred from model size vs peak VRAM",
        "mtp": false,
        "threads": null,
        "recipe": null
      },
      "contexts": [
        {
          "tokens": 4096,
          "label": "short",
          "prefill_tok_s": 2710.8,
          "decode_tok_s": 109.73,
          "vram_gb": 24.68
        },
        {
          "tokens": 32768,
          "label": "32K",
          "prefill_tok_s": 1902.6,
          "decode_tok_s": 96.21,
          "vram_gb": 25.39
        },
        {
          "tokens": 131072,
          "label": "128K",
          "prefill_tok_s": 940.3,
          "decode_tok_s": 69.64,
          "vram_gb": 27.43
        },
        {
          "tokens": 262144,
          "label": "256K",
          "prefill_tok_s": 541.7,
          "decode_tok_s": 40.54,
          "vram_gb": 29.15
        }
      ]
    },
    {
      "id": "local-desktop-qwen36-q4-k-m",
      "trust": "verified",
      "hardware": {
        "host": "local-desktop",
        "gpu": "Radeon AI PRO R9700",
        "gpu_id": "r9700",
        "gfx": "gfx1201",
        "gpu_count": 1,
        "gpus_used": 1,
        "vram_gb_each": 32,
        "ram_gb": 256,
        "fingerprint": "local-desktop|r9700|gfx1201|32|256|1"
      },
      "pins": {
        "model": "Qwen3.6-35B-A3B",
        "official": "bartowski/Qwen_Qwen3.6-35B-A3B",
        "gguf": "bartowski/Qwen_Qwen3.6-35B-A3B-GGUF",
        "quant": "Q4_K_M",
        "backend": "llama.cpp",
        "runtime": "Vulkan",
        "kv": "f16",
        "offload": "llama.cpp --fit-target 1024 MiB placed tensors; experts_to_ram inferred from model size vs peak VRAM",
        "mtp": false,
        "threads": null,
        "recipe": null
      },
      "contexts": [
        {
          "tokens": 4096,
          "label": "short",
          "prefill_tok_s": 2733.9,
          "decode_tok_s": 117.69,
          "vram_gb": 21.43
        },
        {
          "tokens": 32768,
          "label": "32K",
          "prefill_tok_s": 1902.1,
          "decode_tok_s": 101.88,
          "vram_gb": 22.17
        },
        {
          "tokens": 131072,
          "label": "128K",
          "prefill_tok_s": 942.1,
          "decode_tok_s": 72.48,
          "vram_gb": 24.27
        },
        {
          "tokens": 262144,
          "label": "256K",
          "prefill_tok_s": 558.5,
          "decode_tok_s": 47.77,
          "vram_gb": 27.03
        }
      ]
    }
  ]
}