Jackrong
MLX Community

Qwen3.5-27B (Claude Opus 4.6 Distill) benchmark on an Apple's logo.M4 Pro · 24 GB

<- Runs

Prompt tokens

40,960

Generation tokens

10,240

Trials passed

10/10

Verified

14.7 tok/s

96.4 tok/s

Peak memory

18.00/24 GB

Barely runs

Trials

Decode / Prefill Speeds

Metadata

metadata.json
{
"runId": "run_561072b4-0f11-4f37-8ba3-0089bdd486c6",
"bundleId": "mlx-qwen3.5-27b-claude-4.6-opus-distilled-ml-5d2b8e",
"status": "verified",
"promptTokens": 40960,
"completionTokens": 10240,
"contextLength": 5120,
"harness": {
"version": "0.1.18",
"gitSha": "unknown"
},
"runtime": {
"name": "mlx_lm",
"version": "0.31.1",
"buildFlags": null
},
"model": {
"displayName": "Qwen3.5-27B (Claude Opus 4.6 Distill)",
"format": "mlx",
"quant": "4bit",
"architecture": "qwen3_5",
"source": "mlx-community/Qwen3.5-27B-Claude-4.6-Opus-Distilled-MLX-4bit",
"fileSizeBytes": 15133043513,
"lab": {
"name": "Jackrong",
"slug": "jackrong"
},
"quantizedBy": {
"name": "MLX Community",
"slug": "mlx-community"
}
},
"device": {
"cpu": "Apple M4 Pro",
"cpuCores": 12,
"gpu": "Apple M4 Pro",
"gpuCores": 16,
"gpuCount": 1,
"ramGb": 24,
"osName": "macOS",
"osVersion": "26.2"
},
"decodeTpsMean": 14.7,
"prefillTpsMean": 96.4,
"ttftP50Ms": 41531.05,
"idleTpsMean": 3888,
"peakRssMb": 18432,
"trialsPassed": 10,
"trialsTotal": 10,
"runnabilityScore": 0.2464,
"bundleSha256": "d5f8ec24e305cfeed098d20108819383a02f46813f0b0c1d6dcc48d19da636e2",
"createdAt": "2026-03-31T19:05:55.471Z"
}