What LLMs Can the Apple M4 16GB Run?

The Apple M4 16GB has 16 GB of unified memory (about 10.7 GB usable by the GPU) and 120 GB/s of bandwidth. It runs 25 of 58 models fully at Q4_K_M with 8K context, up to Qwen 3 14B, and generates about 14 tokens/s on an 8B model.

Apple · Apple Silicon · Released 2024-10 · Updated October 14, 2026

Your GPU or Mac

Quantization

Context length

KV cache precision

25 of 58 models fit

Apple M4 16GB (10.7 GB GPU-usable) at Q4_K_M with 8K context. Largest that fits fully: Qwen 3 14B.

Model	Needs	Status	Speed	Best quant that fits
Ministral 3 14B	9.7 GB	Fits	~8.8 tok/s	Q4_K_M
Mistral Nemo 12B	9.0 GB	Fits	~9.6 tok/s	Q4_K_M
Gemma 3 12B	8.8 GB	Fits	~9.7 tok/s	Q5_K_M
Ministral 3 8B	6.7 GB	Fits	~14 tok/s	Q6_K
Qwen 3 8B	6.6 GB	Fits	~14 tok/s	Q6_K
Qwen 3.5 9B	6.4 GB	Fits	~14 tok/s	Q6_K
Llama 3.1 8B	6.3 GB	Fits	~14 tok/s	Q8_0
DeepSeek R1 Distill Llama 8B	6.3 GB	Fits	~14 tok/s	Q8_0
Mistral 7B v0.3	5.7 GB	Fits	~16 tok/s	Q8_0
Gemma 4 E4B	5.7 GB	Fits	~16 tok/s	Q8_0
Qwen 2.5 7B	5.6 GB	Fits	~16 tok/s	Q8_0
DeepSeek R1 Distill Qwen 7B	5.6 GB	Fits	~16 tok/s	Q8_0
Qwen 3 4B	4.2 GB	Fits	~25 tok/s	FP16 / BF16
Phi-4 Mini	4.1 GB	Fits	~27 tok/s	FP16 / BF16
Gemma 4 E2B	4.0 GB	Fits	~25 tok/s	Q8_0
Gemma 3 4B	3.7 GB	Fits	~27 tok/s	FP16 / BF16
Qwen 3.5 4B	3.7 GB	Fits	~28 tok/s	FP16 / BF16
Ministral 3 3B	3.5 GB	Fits	~31 tok/s	FP16 / BF16
Llama 3.2 3B	3.5 GB	Fits	~32 tok/s	FP16 / BF16
Llama 3.2 1B	1.7 GB	Fits	~87 tok/s	FP16 / BF16
Gemma 3 1B	1.6 GB	Fits	~119 tok/s	FP16 / BF16
Qwen 2.5 14B	10.7 GB	Tight fit	~8.0 tok/s	IQ4_XS
DeepSeek R1 Distill Qwen 14B	10.7 GB	Tight fit	~8.0 tok/s	IQ4_XS
Phi-4 14B	10.6 GB	Tight fit	~8.0 tok/s	IQ4_XS
Qwen 3 14B	10.5 GB	Tight fit	~8.1 tok/s	IQ4_XS
Kimi K2.6	566 GB	Too small	–	–
DeepSeek R1 671B	380 GB	Too small	–	–
Llama 3.1 405B	234 GB	Too small	–	–
Qwen 3.5 397B-A17B	226 GB	Too small	–	–
DeepSeek V4 Flash	161 GB	Too small	–	–
Qwen 3 235B-A22B	135 GB	Too small	–	–
MiniMax-M2	133 GB	Too small	–	–
Qwen 3.5 122B-A10B	70.2 GB	Too small	–	–
Mistral Small 4 119B	68.2 GB	Too small	–	–
gpt-oss-120b	67.2 GB	Too small	–	–
Llama 4 Scout	64.0 GB	Too small	–	–
GLM-4.5-Air	62.1 GB	Too small	–	–
Qwen3-Next 80B-A3B	46.3 GB	Too small	–	–
Qwen 2.5 72B	44.5 GB	Too small	–	–
Llama 3.3 70B	43.2 GB	Too small	–	–
DeepSeek R1 Distill Llama 70B	43.2 GB	Too small	–	–
Mixtral 8x7B	28.0 GB	Too small	–	–
Qwen 2.5 32B	21.4 GB	Too small	–	–
Qwen 2.5 Coder 32B	21.4 GB	Too small	–	–
DeepSeek R1 Distill Qwen 32B	21.4 GB	Too small	–	–
Qwen 3 32B	21.4 GB	Too small	–	–
Qwen 3.5 35B-A3B	21.0 GB	Too small	–	–
Qwen 3.6 35B-A3B	21.0 GB	Too small	–	–
Gemma 4 31B	19.8 GB	Too small	–	–
Qwen 3 30B-A3B	18.8 GB	Too small	–	–
Qwen3-Coder 30B-A3B	18.8 GB	Too small	–	–
Gemma 3 27B	17.6 GB	Too small	–	–
Qwen 3.5 27B	16.9 GB	Too small	–	–
Qwen 3.6 27B	16.9 GB	Too small	–	–
Qwen 3.8 27B	16.9 GB	Too small	–	–
Mistral Small 3.1 24B	15.6 GB	Too small	–	–
Gemma 4 26B-A4B	15.6 GB	Too small	–	–
gpt-oss-20b	12.9 GB	Too small	–	Q2_K

Apple M4 16GB for local AI

The base Mac mini and MacBook. macOS lets the GPU use about two thirds of 16 GB by default, so plan for about 10.7 GB. Good for 7B to 8B models at modest speed.

Among the 2 devices we track with about 11 GB of usable memory, the Apple M4 16GB is the slowest at generating text, because generation speed follows memory bandwidth. Others in the same class: RTX 2080 Ti 11GB. They all hold the same models; the faster ones simply answer sooner.

Our picks for the Apple M4 16GB

The model we would run first for each job, chosen from our recommendations by what fits fully with an 8K context. "Best quant" is the highest quality that fits; "max context" is at Q4_K_M or the native format.

Task	Model	Best quant	Max context	Speed
Best all-rounder	Qwen 3.5 9B	Q6_K	125K	~14 tok/s
Coding	Qwen 3.5 9B	Q6_K	125K	~14 tok/s
Fastest replies	Qwen 3.5 4B	FP16 / BF16	211K	~28 tok/s
Long documents (64K+)	Qwen 3.5 9B	Q6_K	125K	~14 tok/s
Images in, text out	Ministral 3 14B	Q4_K_M	10K	~8.8 tok/s
Image generation	FLUX.1 [dev]	GGUF Q5_K_S	–	–

Spec	Value
Memory	16 GB unified (10.7 GB GPU-usable by default)
Memory bandwidth	120 GB/s
Architecture	Apple Silicon
Released	2024-10
Llama 3.1 8B Q4_K_M speed	~14 tok/s
Largest model that fits	Qwen 3 14B

Image and video models on the Apple M4 16GB

Peak VRAM with model swapping (ComfyUI default on smaller cards).

Model	FP8	FP8 fit	GGUF Q4	Q4 fit
Stable Diffusion 1.5	1.8 GB	Fits	1.7 GB	Fits
SDXL 1.0	4.1 GB	Fits	4.1 GB	Fits
Stable Diffusion 3.5 Medium	6.4 GB	Fits	6.4 GB	Fits
Stable Diffusion 3.5 Large	8.6 GB	Fits	6.4 GB	Fits
FLUX.1 [dev]	12.1 GB	Too small	7.3 GB	Fits
FLUX.1 [schnell]	12.1 GB	Too small	7.3 GB	Fits
Qwen-Image	20.4 GB	Too small	12.3 GB	Too small
Wan 2.2 A14B (T2V / I2V)	18.8 GB	Too small	13.2 GB	Too small
Wan 2.2 TI2V 5B	8.0 GB	Fits	6.5 GB	Fits
HunyuanVideo	18.3 GB	Too small	13.1 GB	Too small
FLUX.1 Kontext [dev]	12.3 GB	Too small	7.5 GB	Fits
FLUX.2 [dev]	31.0 GB	Too small	23.6 GB	Too small
Z-Image Turbo	6.8 GB	Fits	4.9 GB	Fits
Qwen-Image-Edit 2509	20.5 GB	Too small	12.4 GB	Too small
LTX-2	19.2 GB	Too small	12.6 GB	Too small

Upgrade path

The next step up is the Apple M4 Pro 24GB with 24 GB and 127% more bandwidth.

Apple M4 16GB for LLMs: common questions

Can the Apple M4 16GB run Llama 3.3 70B?

No. It needs about 43.2 GB at Q4_K_M, far more than the 10.7 GB available.

Can the Apple M4 16GB run Qwen 3 32B?

No. It needs about 21.4 GB at Q4_K_M, far more than the 10.7 GB available.

Can the Apple M4 16GB run Qwen 3 14B?

Yes. Qwen 3 14B needs about 10.5 GB at Q4_K_M with 8K context, and the best quality that fits is IQ4_XS.

How fast is the Apple M4 16GB for LLMs?

About 14 tokens per second for an 8B model at Q4_K_M, based on its 120 GB/s memory bandwidth. Larger models scale down roughly in proportion to their size.

How much of the 16 GB can the GPU use?

About 10.7 GB by default. You can raise it with sudo sysctl iogpu.wired_limit_mb=<MB> (resets on reboot), leaving at least 8 GB for macOS.