Arao AI
Back to search results

Qwen3-4B-Thinking-2507 · Q5_K_S

Start a discussion
Apple M5 Max128GBOllama v0.54L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (Q5_K_S) on Apple M5 Max with Ollama, contributed by 1 independent source platforms; metrics are averages of published measurements.

192.3 tok/s

Decode

Decode speed

1,031.35 tok/s

Prefill

Prefill speed

0.19 s

TTFT

Time to first token

9.66 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. Ollama · Q5_K_S (current)Decode 192.3 · Prefill 1031.35 ·
  2. Strata · Q5_K_M Decode 281.24 · Prefill 1191.19 ·
  3. vLLM · Q4_K_S Decode 240.97 · Prefill 1176.03 ·
  4. Strata · IQ4_XS Decode 348.81 · Prefill 1453.67 ·

Core figures

Decode (avg)
192.3 tok/s
Prefill (avg)
1,031.35 tok/s
TTFT (avg)
0.19 s
VRAM (avg)
9.66 GB
MTP acceptance rate
—
TTFB
1.54 GB
Power draw
264.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B-Thinking-2507
Quantization
Q5_K_S
Framework
Ollama
Version
v0.54
Context length
16384 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)
9.66 GB
OS
Ubuntu 22.04
Driver
560.94
CUDA
12.6
Power draw
264.6 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.54 · Ubuntu 22.04 · CUDA 12.6 · 16384 ctx

    192.3 tok/s

    Decode

    1,031.35 tok/s

    Prefill

    0.19 s

    TTFT

    9.66 GB

    VRAM

    —

    MTP

    264.6 W

    Power draw

    L2 batch seed 2026-10-16: Ollama on hw_id=28 running model_id=57 at Q5_K_S