M2 — benchmark record

M2 local LLM benchmarks across 8 GB, 16 GB, 24 GB RAM tiers on Apple Silicon Mac. 8 published rows across 8 models with explicit evidence state and RAM-tier comparison. Peak published speed is 72.0 tok/s.

8Benchmark rows

8Models tested

3RAM configurations

72.0Fastest 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.

8 GBM2 (8 GB)3 rows · 3 models72.0 tok/s peak 16 GBM2 (16 GB)4 rows · 4 models58.0 tok/s peak 24 GBM2 (24 GB)1 rows · 1 models6.0 tok/s peak

All benchmark rows — M2

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
M2 (8 GB)	Phi-4 Mini Instruct 3.8B	Q4_K - Medium	—	72.0 tok/s	—	Ollama	ref
M2 (16 GB)	DeepSeek R1 Distill Llama 8B	Q4_K - Medium	—	58.0 tok/s	—	Ollama	ref
M2 (16 GB)	Qwen 3 8B	Q4_K - Medium	—	55.0 tok/s	—	Ollama	ref
M2 (8 GB)	Qwen 3 4B	Q4_K - Medium	—	52.0 tok/s	—	Ollama	ref
M2 (8 GB)	Llama 3.1 8B	Q4_K - Medium	—	48.0 tok/s	—	Ollama	ref
M2 (16 GB)	Gemma 3 12B	Q4_K - Medium	—	32.0 tok/s	—	Ollama	ref
M2 (16 GB)	Phi-4 14B	Q4_K - Medium	—	28.0 tok/s	—	MLX	ref
M2 (24 GB)	Devstral Small 1.1	4bit	—	6.0 tok/s	—	LM Studio	ref

Models tested on M2

DeepSeek R1 Distill Llama 8B Devstral Small 1.1 Gemma 3 12B Llama 3.1 8B Phi-4 14B Phi-4 Mini Instruct 3.8B Qwen 3 4B Qwen 3 8B

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 →