M4 — benchmark record

M4 local LLM benchmarks across 16 GB, 32 GB RAM tiers on Apple Silicon Mac. 15 published rows across 10 models with explicit evidence state and RAM-tier comparison. Peak published speed is 92.0 tok/s.

15Benchmark rows

10Models tested

2RAM configurations

92.0Fastest avg tok/s

RAM configurations

Each configuration differs only in unified memory. More RAM = larger models fit. Throughput is similar across RAM tiers at the same model size.

16 GBM4 (16 GB)14 rows · 9 models92.0 tok/s peak 32 GBM4 (32 GB)1 rows · 1 models5.7 tok/s peak

All benchmark rows — M4

Sorted by avg tok/s descending. Click source badge to see original measurement.

Chip (RAM)	Model	Quant	RAM req.	Avg tok/s	Prompt tok/s	Runtime	Source
M4 (16 GB)	Qwen3.5-4B	Q4_K - Medium	—	92.0 tok/s	—	Ollama	ref
M4 (16 GB)	DeepSeek R1 Distill Llama 8B	Q4_K - Medium	—	78.0 tok/s	—	MLX	ref
M4 (16 GB)	Llama 3.1 8B	Q4_K - Medium	—	75.0 tok/s	—	Ollama	ref
M4 (16 GB)	Qwen3.5-9B	Q4_K - Medium	—	72.0 tok/s	—	LM Studio	ref
M4 (16 GB)	Ministral 3 8B	Q4_K - Medium	—	72.0 tok/s	—	MLX	ref
M4 (16 GB)	Phi-4 14B	Q4_K - Medium	—	38.0 tok/s	—	Ollama	ref
M4 (32 GB)	Gemma 3 27B	Q4_0	—	5.7 tok/s	47.5 tok/s	llama.cpp	ref
M4 (16 GB)	Qwen3.5-9B	Q4_0	—	4.1 tok/s	—	llama.cpp	ref
M4 (16 GB)	Devstral Small 2 24B	Q4_0	—	3.4 tok/s	—	llama.cpp	ref
M4 (16 GB)	Qwen3.5-9B	Q4_K - Small	—	3.1 tok/s	—	llama.cpp	ref
M4 (16 GB)	Qwen3.5-9B	Q6_K	—	2.2 tok/s	—	llama.cpp	ref
M4 (16 GB)	Qwen3.5-35B-A3B	Q4_K - Medium	—	1.3 tok/s	—	llama.cpp	ref
M4 (16 GB)	Devstral Small 2 24B	Q4_1	—	0.1 tok/s	—	llama.cpp	ref
M4 (16 GB)	Devstral Small 2 24B	Q4_K - Medium	—	0.0 tok/s	—	llama.cpp	ref
M4 (16 GB)	Qwen3.5-27B	Q4_K - Medium	—	0.0 tok/s	—	llama.cpp	ref

Models tested on M4

DeepSeek R1 Distill Llama 8B Devstral Small 2 24B Gemma 3 27B Llama 3.1 8B Ministral 3 8B Phi-4 14B Qwen3.5-4B Qwen3.5-9B Qwen3.5-27B Qwen3.5-35B-A3B

Data

benchmarks.json — full dataset · chips.json — chip summaries · benchmarks.csv — CSV export

Data from in-house lab measurements plus community-published benchmarks. See all chip families →