Arao AI
Back to search results

Qwen3-Coder-Next · Q4_K_S

Start a discussion
Apple M2 Ultrallama.cpp v0.54L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-Coder-Next (Q4_K_S) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

104.58 tok/s

Decode

Decode speed

471.99 tok/s

Prefill

Prefill speed

1.13 s

TTFT

Time to first token

1.05 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. llama.cpp · Q4_K_S (current)Decode 104.58 · Prefill 471.99 ·
  2. vLLM · IQ1_S Decode 161.77 · Prefill 872.93 ·
  3. vLLM · Q5_K_S Decode 96.07 · Prefill 412.07 ·

Core figures

Decode (avg)
104.58 tok/s
Prefill (avg)
471.99 tok/s
TTFT (avg)
1.13 s
VRAM (avg)
1.05 GB
MTP acceptance rate
—
TTFB
0.84 GB
Power draw
22.3 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-Coder-Next
Quantization
Q4_K_S
Framework
llama.cpp
Version
v0.54
Context length
16384 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 M2 Ultra
Measured VRAM (avg)
1.05 GB
OS
macOS 15.0
Driver
Apple M2 Ultra
CUDA
N/A
Power draw
22.3 W

Sources & evidence

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

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

    v0.54 · macOS 15.0 · CUDA N/A · 16384 ctx

    104.58 tok/s

    Decode

    471.99 tok/s

    Prefill

    1.13 s

    TTFT

    1.05 GB

    VRAM

    —

    MTP

    22.3 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=103 running model_id=58 at Q4_K_S