Arao AI
Back to search results

Qwen3.5-4B · IQ1_S

Start a discussion
Apple M5 Ultra512GBllama.cpp v0.37L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ1_S) on Apple M5 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

190.95 tok/s

Decode

Decode speed

992.52 tok/s

Prefill

Prefill speed

2.15 s

TTFT

Time to first token

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

Core figures

Decode (avg)
190.95 tok/s
Prefill (avg)
992.52 tok/s
TTFT (avg)
2.15 s
VRAM (avg)
5.63 GB
MTP acceptance rate
—
TTFB
1.53 GB
Power draw
98.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ1_S
Framework
llama.cpp
Version
v0.37
Context length
32768 tokens
Batch size
512
GPU layers
33
Flash Attention
Off

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)
5.63 GB
OS
macOS 26.1
Power draw
98.7 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

    v0.37 · macOS 26.1 · 32768 ctx

    190.95 tok/s

    Decode

    992.52 tok/s

    Prefill

    2.15 s

    TTFT

    5.63 GB

    VRAM

    —

    MTP

    98.7 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=29 running model_id=48 at IQ1_S