Arao AI
Back to search results

Raon-1-mini-preview · IQ3_S

Start a discussion
Apple M2 Ultrallama.cpp v0.39L2 Cross-framework verified

This page aggregates 1 real-world runs of Raon-1-mini-preview (IQ3_S) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

130 tok/s

Decode

Decode speed

684.42 tok/s

Prefill

Prefill speed

1.54 s

TTFT

Time to first token

42.02 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

13 days ago

Last verified

Performance

  1. llama.cpp · IQ2_M Decode 155.94 · Prefill 578.17 ·
  2. llama.cpp · IQ3_S (current)Decode 130 · Prefill 684.42 ·
  3. Ollama · Q4_K_M Decode 110.45 · Prefill 501.39 ·

Core figures

Decode (avg)
130 tok/s
Prefill (avg)
684.42 tok/s
TTFT (avg)
1.54 s
VRAM (avg)
42.02 GB
MTP acceptance rate
—
TTFB
1.04 GB
Power draw
145.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Raon-1-mini-preview
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.39
Context length
16384 tokens
Batch size
512
GPU layers
20
Flash Attention
On

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)
42.02 GB
OS
macOS 26.1
Power draw
145.9 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

    v0.39 · macOS 26.1 · 16384 ctx

    130 tok/s

    Decode

    684.42 tok/s

    Prefill

    1.54 s

    TTFT

    42.02 GB

    VRAM

    —

    MTP

    145.9 W

    Power draw

    L2 batch seed 2026-10-22: llama.cpp on hw_id=103 running model_id=93 at IQ3_S