Llama 3.2 3B Instruct benchmark on an M5 Pro · 48 GB

{
  "runId": "run_4cd9d1b1-8bd0-4ab4-b6cc-eb0d03e0bd14",
  "bundleId": "mlx-llama-3.2-3b-instruct-4bit-278877",
  "status": "verified",
  "promptTokens": 40960,
  "completionTokens": 10240,
  "contextLength": 5120,
  "harness": {
    "version": "0.1.20",
    "gitSha": "unknown"
  },
  "runtime": {
    "name": "mlx_lm",
    "version": "0.31.2",
    "buildFlags": null
  },
  "model": {
    "displayName": "Llama 3.2 3B Instruct",
    "format": "mlx",
    "quant": "4bit",
    "architecture": "llama",
    "source": "mlx-community/Llama-3.2-3B-Instruct-4bit",
    "fileSizeBytes": 1807496278,
    "lab": {
      "name": "Meta",
      "slug": "meta"
    },
    "quantizedBy": {
      "name": "MLX Community",
      "slug": "mlx-community"
    }
  },
  "device": {
    "cpu": "Apple M5 Pro",
    "cpuCores": 15,
    "gpu": "Apple M5 Pro",
    "gpuCores": 16,
    "gpuCount": 1,
    "ramGb": 48,
    "osName": "macOS",
    "osVersion": "26.4.1"
  },
  "decodeTpsMean": 106.3,
  "prefillTpsMean": 3225,
  "ttftP50Ms": 1313.94,
  "idleTpsMean": 1566,
  "peakRssMb": 3838,
  "trialsPassed": 10,
  "trialsTotal": 10,
  "runnabilityScore": 0.9471602957589286,
  "bundleSha256": "f79fb464c2b90ad9dd50a02662da147bead2d34b2ba6b506831a667efb14b160",
  "createdAt": "2026-04-13T18:25:57.493Z"
}