What LLMs Can I Run on a Mac (M2, 16 GB)?

16 GB of unified memory, of which about 10.7 GB is usable for a model. That runs 11 of the 20 models here , up to DeepSeek-R1 Distill Qwen 14B.

Largest model that fits: DeepSeek-R1 Distill Qwen 14B at Q4_K_M, needing 10.0 GB and leaving 0.7 GB spare. Its context can run to 12K tokens before memory runs out.

Every model, on this hardware

Which models run on a Mac (M2, 16 GB)
Model	Size	Needs	Runs	Max context
Llama 3.2 1B	1.2B	1.7 GB	Yes	128K
Llama 3.2 3B	3.2B	3.4 GB	Yes	75K
Llama 3.1 8B	8.0B	6.0 GB	Yes	46K
Llama 3.3 70B	70.5B	40.3 GB	No	—
Qwen2.5 3B	3.1B	2.7 GB	Yes	32K
Qwen2.5 7B	7.6B	5.2 GB	Yes	32K
Qwen2.5 14B	14.8B	10.0 GB	Yes	12K
Qwen2.5 32B	32.8B	20.0 GB	No	—
Qwen2.5 72B	72.7B	41.4 GB	No	—
Qwen2.5 Coder 7B	7.6B	5.2 GB	Yes	32K
Qwen2.5 Coder 32B	32.8B	20.0 GB	No	—
Qwen3 32B	32.8B	20.0 GB	No	—
Mistral 7B v0.3	7.3B	5.6 GB	Yes	32K
Gemma 2 9B	9.2B	8.3 GB	Yes	8K
Gemma 2 27B	27.2B	17.9 GB	No	—
Phi-4	14.7B	10.0 GB	Yes	11K
Qwen3 30B-A3B	30.5B	17.5 GB	No	—
Qwen3 235B-A22B	235.1B	125.4 GB	No	—
Mixtral 8x7B	46.7B	26.3 GB	No	—
DeepSeek-R1 Distill Qwen 14B	14.8B	10.0 GB	Yes	12K

Q4_K_M at 8K context. "Max context" is the longest context that still fits once the weights are loaded.

Unified memory changes the arithmetic

A Mac shares one pool between the CPU and GPU, so a laptop with 16 GB can hold a model that no consumer graphics card can — the largest of those tops out at 32 GB. The trade is bandwidth: generation is slower than a discrete card of comparable capacity. For a model that would otherwise not run at all, slower is the better problem.

The other half is context. Weights are a fixed cost, but the key-value cache grows with every token, and on a long conversation it can outgrow the weights. That is why the table above gives a maximum context per model rather than a yes or no — plenty of models load fine on this hardware and then run out of room mid-conversation.

Frequently asked questions

What LLMs can a Mac (M2, 16 GB) run?

11 of the 20 models on this site, at Q4_K_M with a 8K context. The largest is DeepSeek-R1 Distill Qwen 14B at 14.8B parameters, needing 10.0 GB of the roughly 10.7 GB usable.

Why is usable memory less than the 16 GB on the box?

Because the memory is shared with everything else the machine is doing. macOS caps what it will hand to one process, so of the 16 GB here roughly 10.7 GB is available to a model. That is still the advantage of unified memory — a laptop holding a model no consumer graphics card can — but the headline figure is not all yours.

Does a bigger model at a rougher quantisation beat a smaller one?

Usually, yes. A 32B model at Q3_K_M generally outperforms a 7B at Q8_0, even though the smaller model is barely quantised at all. The practical rule is to fit the largest model you can and then step the quantisation down only as far as you must — which is what the "best format" column above is showing.

How accurate are these figures?

They count weights, the key-value cache at the chosen context, and a fixed runtime allowance. Real usage moves with the runtime, the batch size and whether flash attention is enabled. Treat anything within a gigabyte of capacity as "probably not" rather than "just fits" — which is why the table says "only just" rather than "yes" past 95% of usable memory.

Manufacturer specifications for VRAM and unified memory. Model config.json files published on Hugging Face.

An estimate of memory, not a benchmark. It counts model weights, the key-value cache at the context you choose, and a fixed runtime allowance. Actual usage moves with the runtime, the batch size, whether flash attention is on, and how much the operating system has already taken from a shared memory pool. Treat a result within a gigabyte of your card's capacity as 'probably not' rather than 'just fits'.

Data on this page last verified 2 September 2026.