M3 — benchmark record

M3 local LLM benchmarks across 16 GB RAM tiers on Apple Silicon Mac. 9 published rows across 9 models with explicit evidence state and RAM-tier comparison. Peak published speed is 95.0 tok/s.

9Benchmark rows

9Models tested

1RAM configurations

95.0Fastest 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 GBM3 (16 GB)9 rows · 9 models95.0 tok/s peak

All benchmark rows — M3

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
M3 (16 GB)	Phi-4 Mini Instruct 3.8B	Q4_K - Medium	—	95.0 tok/s	—	MLX	ref
M3 (16 GB)	Gemma 4 E2B	Q4_K - Medium	—	82.0 tok/s	—	Ollama	ref
M3 (16 GB)	Gemma 4 E4B	Q4_K - Medium	—	62.0 tok/s	—	Ollama	ref
M3 (16 GB)	Mistral 7B v0.3	Q4_K - Medium	—	62.0 tok/s	—	Ollama	ref
M3 (16 GB)	Qwen3.5-9B	Q4_K - Medium	—	58.0 tok/s	—	Ollama	ref
M3 (16 GB)	Ministral 3 8B	Q4_K - Medium	—	55.0 tok/s	—	LM Studio	ref
M3 (16 GB)	Gemma 3 12B	Q4_K - Medium	—	38.0 tok/s	—	Ollama	ref
M3 (16 GB)	Qwen 3 14B	Q4_K - Medium	—	30.0 tok/s	—	Ollama	ref
M3 (16 GB)	Ministral 3 14B	Q4_K - Medium	—	30.0 tok/s	—	LM Studio	ref

Models tested on M3

Gemma 3 12B Gemma 4 E2B Gemma 4 E4B Ministral 3 8B Ministral 3 14B Mistral 7B v0.3 Phi-4 Mini Instruct 3.8B Qwen 3 14B Qwen3.5-9B

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 →