Arao AI
Back to search results

Ministral-3-8B-Instruct-2512 · Q5_K_M

Start a discussion
Apple M5 Ultra512GBllama.cpp v0.21L2 Cross-framework verified

This page aggregates 1 real-world runs of Ministral-3-8B-Instruct-2512 (Q5_K_M) on Apple M5 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

84.63 tok/s

Decode

Decode speed

432.9 tok/s

Prefill

Prefill speed

0.78 s

TTFT

Time to first token

17.72 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 · Q5_K_M (current)Decode 84.63 · Prefill 432.9 ·
  2. Ollama · Q3_K_S Decode 125.07 · Prefill 471.87 ·

Core figures

Decode (avg)
84.63 tok/s
Prefill (avg)
432.9 tok/s
TTFT (avg)
0.78 s
VRAM (avg)
17.72 GB
MTP acceptance rate
—
TTFB
0.68 GB
Power draw
93.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Ministral-3-8B-Instruct-2512
Quantization
Q5_K_M
Framework
llama.cpp
Version
v0.21
Context length
65536 tokens
Batch size
256
GPU layers
99
Flash Attention
On

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)
17.72 GB
OS
macOS 26.1
Power draw
93.9 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

    v0.21 · macOS 26.1 · 65536 ctx

    84.63 tok/s

    Decode

    432.9 tok/s

    Prefill

    0.78 s

    TTFT

    17.72 GB

    VRAM

    —

    MTP

    93.9 W

    Power draw

    L2 batch seed 2026-10-24: llama.cpp on hw_id=29 running model_id=74 at Q5_K_M