Arao AI
Back to search results

Qwen3.5-4B · IQ2_M

Start a discussion
Apple M3 Max36GBllama.cpp v0.49L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ2_M) on Apple M3 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

188.61 tok/s

Decode

Decode speed

924.46 tok/s

Prefill

Prefill speed

1.9 s

TTFT

Time to first token

6.47 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ2_M (current)Decode 188.61 · Prefill 924.46 ·
  2. llama.cpp · Q6_K Decode 107.73 · Prefill 580.34 ·
  3. Ollama · Q4_K_S Decode 130.85 · Prefill 560.14 ·
  4. Strata · Q3_K_M Decode 237.4 · Prefill 1147.78 ·
  5. Strata · Q6_K Decode 145.1 · Prefill 751.14 ·

Core figures

Decode (avg)
188.61 tok/s
Prefill (avg)
924.46 tok/s
TTFT (avg)
1.9 s
VRAM (avg)
6.47 GB
MTP acceptance rate
—
TTFB
1.51 GB
Power draw
83.3 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ2_M
Framework
llama.cpp
Version
v0.49
Context length
32768 tokens
Batch size
256
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 M3 Max
Nominal VRAM
36 GB
Measured VRAM (avg)
6.47 GB
OS
macOS 15.0
Driver
Apple M3 Max
CUDA
N/A
Power draw
83.3 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

    v0.49 · macOS 15.0 · CUDA N/A · 32768 ctx

    188.61 tok/s

    Decode

    924.46 tok/s

    Prefill

    1.9 s

    TTFT

    6.47 GB

    VRAM

    —

    MTP

    83.3 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=4 running model_id=48 at IQ2_M