Arao AI
Back to search results

Ministral-3-8B-Instruct-2512 · IQ4_XS

Start a discussion
Apple M5 Max128GBllama.cpp v0.27L2 Cross-framework verified

This page aggregates 1 real-world runs of Ministral-3-8B-Instruct-2512 (IQ4_XS) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

75.66 tok/s

Decode

Decode speed

303.23 tok/s

Prefill

Prefill speed

2.22 s

TTFT

Time to first token

14.53 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

12 days ago

Last verified

Performance

  1. llama.cpp · IQ4_XS (current)Decode 75.66 · Prefill 303.23 ·
  2. Ollama · IQ4_XS Decode 67.21 · Prefill 271.49 ·

Core figures

Decode (avg)
75.66 tok/s
Prefill (avg)
303.23 tok/s
TTFT (avg)
2.22 s
VRAM (avg)
14.53 GB
MTP acceptance rate
—
TTFB
0.61 GB
Power draw
54.1 W

Configuration

Member-level fields are taken from the most recent run

Model
Ministral-3-8B-Instruct-2512
Quantization
IQ4_XS
Framework
llama.cpp
Version
v0.27
Context length
4096 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)
14.53 GB
OS
macOS 26.1
Power draw
54.1 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-24

    v0.27 · macOS 26.1 · 4096 ctx

    75.66 tok/s

    Decode

    303.23 tok/s

    Prefill

    2.22 s

    TTFT

    14.53 GB

    VRAM

    —

    MTP

    54.1 W

    Power draw

    L2 batch seed 2026-10-24: llama.cpp on hw_id=28 running model_id=74 at IQ4_XS