Arao AI
Back to search results

Qwen3.5-4B · Q3_K_L

Start a discussion
Apple M5 Ultra512GBOllama v0.45L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (Q3_K_L) on Apple M5 Ultra with Ollama, contributed by 1 independent source platforms; metrics are averages of published measurements.

144 tok/s

Decode

Decode speed

660.08 tok/s

Prefill

Prefill speed

2.32 s

TTFT

Time to first token

8.15 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 · IQ1_S Decode 190.95 · Prefill 992.52 ·
  2. Ollama · Q3_K_L (current)Decode 144 · Prefill 660.08 ·

Core figures

Decode (avg)
144 tok/s
Prefill (avg)
660.08 tok/s
TTFT (avg)
2.32 s
VRAM (avg)
8.15 GB
MTP acceptance rate
—
TTFB
1.15 GB
Power draw
127.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
Q3_K_L
Framework
Ollama
Version
v0.45
Context length
2048 tokens
Batch size
512
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 M5 Ultra
Nominal VRAM
512 GB
Measured VRAM (avg)
8.15 GB
OS
macOS 26.1
Power draw
127.8 W

Sources & evidence

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

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

    v0.45 · macOS 26.1 · 2048 ctx

    144 tok/s

    Decode

    660.08 tok/s

    Prefill

    2.32 s

    TTFT

    8.15 GB

    VRAM

    —

    MTP

    127.8 W

    Power draw

    L2 batch seed 2026-10-19: Ollama on hw_id=29 running model_id=48 at Q3_K_L