Arao AI
Back to search results

Qwen3.8-27B · Q6_K

Start a discussion
Apple M5 Max128GBllama.cpp v0.48L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-27B (Q6_K) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

30.73 tok/s

Decode

Decode speed

132.88 tok/s

Prefill

Prefill speed

0.81 s

TTFT

Time to first token

58.8 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 · Q3_K_L Decode 42.74 · Prefill 158.84 ·
  2. llama.cpp · Q6_K (current)Decode 30.73 · Prefill 132.88 ·
  3. Ollama · IQ3_XS Decode 46.88 · Prefill 178.13 ·
  4. llama.cpp · Q5_K_M Decode 23.9 · Prefill 179.3 ·

Core figures

Decode (avg)
30.73 tok/s
Prefill (avg)
132.88 tok/s
TTFT (avg)
0.81 s
VRAM (avg)
58.8 GB
MTP acceptance rate
—
TTFB
0.25 GB
Power draw
40.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-27B
Quantization
Q6_K
Framework
llama.cpp
Version
v0.48
Context length
4096 tokens
Batch size
256
GPU layers
28
Flash Attention
On

Hardware

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

GPU
Apple M5 Max
Nominal VRAM
128 GB
Measured VRAM (avg)
58.8 GB
OS
macOS 26.1
Power draw
40.7 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

    v0.48 · macOS 26.1 · 4096 ctx

    30.73 tok/s

    Decode

    132.88 tok/s

    Prefill

    0.81 s

    TTFT

    58.8 GB

    VRAM

    —

    MTP

    40.7 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=28 running model_id=5 at Q6_K