Back to search results
Qwen3.5-4B · IQ3_XS
Apple M2 Ultrallama.cpp v0.42L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3.5-4B (IQ3_XS) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
156.36 tok/s
Decode
Decode speed
856.45 tok/s
Prefill
Prefill speed
2.3 s
TTFT
Time to first token
7.31 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
16 days ago
Last verified
Performance
- llama.cpp · IQ3_XS (current)Decode 156.36 · Prefill 856.45 ·
- llama.cpp · Q3_K_L Decode 147.11 · Prefill 678.92 ·
- Ollama · IQ2_M Decode 176 · Prefill 698.29 ·
- vLLM · Q3_K_L Decode 164.19 · Prefill 535.64 ·
- vLLM · Q3_K_S Decode 169.83 · Prefill 644.07 ·
- Ollama · FP16 Decode 60.62 · Prefill 204.54 ·
Core figures
- Decode (avg)
- 156.36 tok/s
- Prefill (avg)
- 856.45 tok/s
- TTFT (avg)
- 2.3 s
- VRAM (avg)
- 7.31 GB
- MTP acceptance rate
- —
- TTFB
- 1.25 GB
- Power draw
- 100.7 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-4B
- Quantization
- IQ3_XS
- Framework
- llama.cpp
- Version
- v0.42
- 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 M2 Ultra
- Measured VRAM (avg)
- 7.31 GB
- OS
- macOS 26.1
- Power draw
- 100.7 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.42 · macOS 26.1 · 16384 ctx
156.36 tok/s
Decode
856.45 tok/s
Prefill
2.3 s
TTFT
7.31 GB
VRAM
—
MTP
100.7 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=103 running model_id=48 at IQ3_XS