Can the Apple M1 run Llama 3.1 8B?

Yes — the 4-bit build is a 4.9 GB download and fits in 8–16 GB unified memory. Expect roughly ~12 tok/s.

What is the biggest LLM the Apple M1 can run?

Phi-4 Reasoning Vision 15B is the largest model in our catalog that fits (9.1 GB at 4-bit). Expect about ~6 tok/s.

How fast is the Apple M1 for local LLMs?

Token generation is memory-bandwidth bound. At roughly 68 GB/s, the Apple M1 generates about ~12 tok/s on an 8B-class model at 4-bit — speed scales inversely with model size.

Is Apple unified memory good for local LLMs?

Yes — the GPU can use nearly the whole memory pool, so a 32 GB Mac fits models that would need a very expensive dedicated GPU. Bandwidth, not capacity, is usually the speed limit.

← All modelsDEVICE CHECK

What LLMs can the Apple M1 run?

The Apple M1 has 8–16 GB unified memory and roughly 68 GB/s of memory bandwidth. Below is every model in our catalog that fits, with estimated generation speed. Biggest pick: Phi-4 Reasoning Vision 15B at ~6 tok/s.

Specifications

Memory8–16 GB unified memory

Bandwidth~68 GB/s

Memory typeUnified memory

Released2020-11

Models on the Apple M1

38 / 73 models

Model	Download (Q4)	8 GB	16 GB	~Speed
Phi-4 Reasoning Vision 15BMicrosoft	9.1 GB	Won't fit	Runs	~6 tok/s
Qwen 3 14BAlibaba	9.0 GB	Won't fit	Runs	~6 tok/s
DeepSeek R1 14BDeepSeek	9.0 GB	Won't fit	Runs	~6 tok/s
Phi-4 14BMicrosoft	8.9 GB	Won't fit	Runs	~6 tok/s
Ministral 3 14BMistral AI	8.5 GB	Won't fit	Runs	~7 tok/s
OLMo 2 13BAi2	8.3 GB	Won't fit	Runs	~7 tok/s
Gemma 3 12BGoogle	7.4 GB	Won't fit	Runs	~8 tok/s
Mistral Nemo 12BMistral AI	7.4 GB	Won't fit	Runs	~8 tok/s
Gemma 4 12BGoogle	7.3 GB	Won't fit	Runs	~8 tok/s
Mellum 2 12B-A2.5BJetBrains	7.3 GB	Won't fit	Runs	~38 tok/s
Qwen 3.5 9BAlibaba	5.5 GB	Won't fit	Runs	~11 tok/s
GLM-4.6V-FlashZ.ai	5.5 GB	Won't fit	Runs	~11 tok/s
Qwen 2.5 VL 7BAlibaba	5.0 GB	Runs	Runs	~11 tok/s
Qwen 3 8BAlibaba	5.0 GB	Runs	Runs	~12 tok/s
Granite 3.3 8BIBM	5.0 GB	Runs	Runs	~12 tok/s
Llama 3.1 8BMeta	4.9 GB	Runs	Runs	~12 tok/s
DeepSeek R1 8BDeepSeek	4.9 GB	Runs	Runs	~12 tok/s
Gemma 4 E4BGoogle	4.9 GB	Runs	Runs	~21 tok/s
Qwen3-VL 8BAlibaba	4.9 GB	Runs	Runs	~12 tok/s
Ministral 3 8BMistral AI	4.9 GB	Runs	Runs	~12 tok/s
Gemma 3n E4BGoogle	4.7 GB	Runs	Runs	~24 tok/s
Qwen 2.5 Coder 7BAlibaba	4.6 GB	Runs	Runs	~13 tok/s
DeepSeek R1 7BDeepSeek	4.6 GB	Runs	Runs	~13 tok/s
Mistral 7BMistral AI	4.4 GB	Runs	Runs	~13 tok/s
Gemma 4 E2BGoogle	3.1 GB	Runs	Runs	~41 tok/s
Gemma 3 4BGoogle	2.6 GB	Runs	Runs	~22 tok/s
Qwen 3 4BAlibaba	2.4 GB	Runs	Runs	~24 tok/s
Qwen 3.5 4BAlibaba	2.4 GB	Runs	Runs	~24 tok/s
Phi-4 Mini 3.8BMicrosoft	2.3 GB	Runs	Runs	~25 tok/s
Llama 3.2 3BMeta	1.9 GB	Runs	Runs	~30 tok/s
DeepSeek-OCRDeepSeek	1.8 GB	Runs	Runs	~167 tok/s
Ministral 3 3BMistral AI	1.8 GB	Runs	Runs	~32 tok/s
DeepSeek R1 1.5BDeepSeek	1.1 GB	Runs	Runs	~53 tok/s
Qwen 3 1.7BAlibaba	1.0 GB	Runs	Runs	~56 tok/s
SmolLM2 1.7BHugging Face	1.0 GB	Runs	Runs	~56 tok/s
Llama 3.2 1BMeta	0.7 GB	Runs	Runs	~79 tok/s
Gemma 3 1BGoogle	0.6 GB	Runs	Runs	~95 tok/s
Qwen 3 0.6BAlibaba	0.4 GB	Runs	Runs	~159 tok/s

Apple Silicon shares one memory pool between CPU and GPU, so nearly all unified memory is available to the model (minus OS overhead). · Data updated: 2026-06-11 · How we calculate these numbers →