Arao AI
Back to search results

Flint-40B-A2.5B · Q3_K_M

Start a discussion
Apple M5 Max128GBllama.cpp v0.21L2 Cross-framework verified

This page aggregates 1 real-world runs of Flint-40B-A2.5B (Q3_K_M) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

152.51 tok/s

Decode

Decode speed

687.68 tok/s

Prefill

Prefill speed

2.15 s

TTFT

Time to first token

58.94 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

14 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_M (current)Decode 152.51 · Prefill 687.68 ·
  2. Ollama · IQ4_XS Decode 123.41 · Prefill 650.98 ·
  3. Ollama · Q2_K Decode 153.39 · Prefill 803.06 ·

Core figures

Decode (avg)
152.51 tok/s
Prefill (avg)
687.68 tok/s
TTFT (avg)
2.15 s
VRAM (avg)
58.94 GB
MTP acceptance rate
—
TTFB
1.22 GB
Power draw
41.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Flint-40B-A2.5B
Quantization
Q3_K_M
Framework
llama.cpp
Version
v0.21
Context length
16384 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 Max
Nominal VRAM
128 GB
Measured VRAM (avg)
58.94 GB
OS
macOS 26.1
Power draw
41.8 W

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-22

    v0.21 · macOS 26.1 · 16384 ctx

    152.51 tok/s

    Decode

    687.68 tok/s

    Prefill

    2.15 s

    TTFT

    58.94 GB

    VRAM

    —

    MTP

    41.8 W

    Power draw

    L2 batch seed 2026-10-22: llama.cpp on hw_id=28 running model_id=92 at Q3_K_M