Arao AI
Back to search results

Qwen3.5-4B · Q5_K_M

Start a discussion
Apple M4 Max128GBllama.cpp v0.21L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (Q5_K_M) on Apple M4 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

206.25 tok/s

Decode

Decode speed

667.88 tok/s

Prefill

Prefill speed

1.26 s

TTFT

Time to first token

9.91 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · Q5_K_M (current)Decode 206.25 · Prefill 667.88 ·
  2. Ollama · Q3_K_L Decode 222.79 · Prefill 746.94 ·
  3. Ollama · Q4_K_M Decode 207.82 · Prefill 1018.29 ·
  4. Strata · IQ4_XS Decode 330.49 · Prefill 1150.53 ·
  5. Strata · Q2_K Decode 349.71 · Prefill 1794.35 ·

Core figures

Decode (avg)
206.25 tok/s
Prefill (avg)
667.88 tok/s
TTFT (avg)
1.26 s
VRAM (avg)
9.91 GB
MTP acceptance rate
—
TTFB
1.65 GB
Power draw
51 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
Q5_K_M
Framework
llama.cpp
Version
v0.21
Context length
2048 tokens
Batch size
256
GPU layers
33
Flash Attention
On

Hardware

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

GPU
Apple M4 Max
Nominal VRAM
128 GB
Measured VRAM (avg)
9.91 GB
OS
macOS 15.0
Driver
Apple M4 Max
CUDA
N/A
Power draw
51 W

Sources & evidence

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

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

    v0.21 · macOS 15.0 · CUDA N/A · 2048 ctx

    206.25 tok/s

    Decode

    667.88 tok/s

    Prefill

    1.26 s

    TTFT

    9.91 GB

    VRAM

    —

    MTP

    51 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=26 running model_id=48 at Q5_K_M