Qwen3.5-35B-A3B benchmark on an M2 Max · 96 GB

{
  "runId": "run_f3318934-27a2-4602-8542-d98bfd2fa8e7",
  "bundleId": "mlx-qwen3.5-35b-a3b-4bit-e44b1c",
  "status": "verified",
  "promptTokens": 40960,
  "completionTokens": 10240,
  "contextLength": 5120,
  "harness": {
    "version": "0.1.20",
    "gitSha": "88f4261"
  },
  "runtime": {
    "name": "mlx_lm",
    "version": "0.31.1",
    "buildFlags": null
  },
  "model": {
    "displayName": "Qwen3.5-35B-A3B",
    "format": "mlx",
    "quant": "4bit",
    "architecture": "qwen3_5_moe",
    "source": "mlx-community/Qwen3.5-35B-A3B-4bit",
    "fileSizeBytes": 20391679439,
    "lab": {
      "name": "Qwen",
      "slug": "qwen"
    },
    "quantizedBy": {
      "name": "MLX Community",
      "slug": "mlx-community"
    }
  },
  "device": {
    "cpu": "Apple M2 Max",
    "cpuCores": 12,
    "gpu": "Apple M2 Max",
    "gpuCores": 38,
    "gpuCount": 1,
    "ramGb": 96,
    "osName": "macOS",
    "osVersion": "26.3.1"
  },
  "decodeTpsMean": 62.5,
  "prefillTpsMean": 509.3,
  "ttftP50Ms": 8904.29,
  "idleTpsMean": 823,
  "peakRssMb": 21504,
  "trialsPassed": 10,
  "trialsTotal": 10,
  "runnabilityScore": 0.70093,
  "bundleSha256": "d99147091a15ae5f08d893af6291cd5bdd3dd8e4215d029485b850daa70af630",
  "createdAt": "2026-04-09T14:52:36.469Z"
}