Model benchmarks

qwen3.6:27b-mlx local LLM performance

As of July 2026, qwen3.6:27b-mlx runs at up to 24.1 tok/s for local inference (best of 2 community benchmark runs across 2 GPUs).

Ollamanvfp4
ShareRedditX

Model size

27.4B

Peak speed

24.1 tok/s

Average speed

15.9 tok/s

Min memory

19.9 GB

Max context

65,536 tokens

Avg quality

83.3

Benchmark runs

2

GPUs tested

2

Quality by task

Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 2 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.

TaskP5 (low)AvgP95 (high)
Overall82.083.384.5
Agent Workflow77.583.288.9
Code Generation73.074.576.0
Role Play & Narrative94.895.195.4
Research & Analysis79.880.380.7

Performance by hardware and tool

Every hardware/tool/quantization combination qwen3.6:27b-mlx has been benchmarked on, ranked by peak token generation speed. Last updated July 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
Apple M4 MaxOllamanvfp424.1 tok/s24.1 tok/s20.8 GB8,192 tokens81.91
Apple M5 MaxOllamanvfp47.6 tok/s7.6 tok/s19.9 GB65,536 tokens84.61

Frequently asked questions

Is qwen3.6:27b-mlx good for coding?
In our benchmarks, qwen3.6:27b-mlx scores 74.5/100 for coding. It runs at about 15.9 tok/s, so if you want more speed, /home/polaris/models/Tiel-Coder-35B-A3B-MTP-UD-Q6_K_XL.gguf is faster (~35.6 tok/s) and still scores well for coding (84.9/100).
Is qwen3.6:27b-mlx good for agentic (tool-using) tasks?
In our benchmarks, qwen3.6:27b-mlx scores 83.2/100 for agentic workflows. It runs at about 15.9 tok/s, so if you want more speed, muse-glimmer:latest is faster (~34.5 tok/s) and still scores well for agentic workflows (93.0/100).
How fast is qwen3.6:27b-mlx for local inference?
Across 2 community benchmark runs, qwen3.6:27b-mlx reaches up to 24.1 tok/s and averages 15.9 tok/s, with the fastest results on Apple M4 Max.
How much memory does qwen3.6:27b-mlx need?
The leanest observed configuration used about 19.9 GB of memory (quantizations tested: nvfp4).
Which tools have been used to run qwen3.6:27b-mlx?
Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.