Arao AI
Back to search results

Qwen3.6-27B · Q4_K_M

Start a discussion
Apple M2 Ultrallama.cpp v0.22L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.6-27B (Q4_K_M) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

67.38 tok/s

Decode

Decode speed

344.18 tok/s

Prefill

Prefill speed

1.13 s

TTFT

Time to first token

48.59 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. llama.cpp · Q4_K_M (current)Decode 67.38 · Prefill 344.18 ·
  2. Ollama · Q3_K_L Decode 80.85 · Prefill 325.8 ·

Core figures

Decode (avg)
67.38 tok/s
Prefill (avg)
344.18 tok/s
TTFT (avg)
1.13 s
VRAM (avg)
48.59 GB
MTP acceptance rate
—
TTFB
0.54 GB
Power draw
133.3 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.6-27B
Quantization
Q4_K_M
Framework
llama.cpp
Version
v0.22
Context length
2048 tokens
Batch size
256
GPU layers
20
Flash Attention
On

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
Apple M2 Ultra
Measured VRAM (avg)
48.59 GB
OS
macOS 26.1
Power draw
133.3 W

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.22 · macOS 26.1 · 2048 ctx

    67.38 tok/s

    Decode

    344.18 tok/s

    Prefill

    1.13 s

    TTFT

    48.59 GB

    VRAM

    —

    MTP

    133.3 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=103 running model_id=6 at Q4_K_M