What LLMs Can the Apple M4 Pro 48GB Run?

The Apple M4 Pro 48GB has 48 GB of unified memory (about 36.0 GB usable by the GPU) and 273 GB/s of bandwidth. It runs 42 of 58 models fully at Q4_K_M with 8K context, up to Mixtral 8x7B, and generates about 32 tokens/s on an 8B model.

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

42 of 58 models fit

Apple M4 Pro 48GB (36.0 GB GPU-usable) at Q4_K_M with 8K context. Largest that fits fully: Mixtral 8x7B.

ModelNeedsStatusSpeedBest quant that fits
Mixtral 8x7B28.0 GBFits~20 tok/sQ5_K_M
Qwen 2.5 32B21.4 GBFits~8.4 tok/sQ6_K
Qwen 2.5 Coder 32B21.4 GBFits~8.4 tok/sQ6_K
DeepSeek R1 Distill Qwen 32B21.4 GBFits~8.4 tok/sQ6_K
Qwen 3 32B21.4 GBFits~8.4 tok/sQ6_K
Qwen 3.5 35B-A3B21.0 GBFits~66 tok/sQ6_K
Qwen 3.6 35B-A3B21.0 GBFits~66 tok/sQ6_K
Gemma 4 31B19.8 GBFits~8.9 tok/sQ8_0
Qwen 3 30B-A3B18.8 GBFits~55 tok/sQ8_0
Qwen3-Coder 30B-A3B18.8 GBFits~55 tok/sQ8_0
Gemma 3 27B17.6 GBFits~10 tok/sQ8_0
Qwen 3.5 27B16.9 GBFits~11 tok/sQ8_0
Qwen 3.6 27B16.9 GBFits~11 tok/sQ8_0
Qwen 3.8 27B16.9 GBFits~11 tok/sQ8_0
Mistral Small 3.1 24B15.6 GBFits~12 tok/sQ8_0
Gemma 4 26B-A4B15.6 GBFits~57 tok/sQ8_0
gpt-oss-20b12.9 GBFits~65 tok/sMXFP4 (official)
Qwen 2.5 14B10.7 GBFits~18 tok/sFP16 / BF16
DeepSeek R1 Distill Qwen 14B10.7 GBFits~18 tok/sFP16 / BF16
Phi-4 14B10.6 GBFits~18 tok/sFP16 / BF16
Qwen 3 14B10.5 GBFits~18 tok/sFP16 / BF16
Ministral 3 14B9.7 GBFits~20 tok/sFP16 / BF16
Mistral Nemo 12B9.0 GBFits~22 tok/sFP16 / BF16
Gemma 3 12B8.8 GBFits~22 tok/sFP16 / BF16
Ministral 3 8B6.7 GBFits~31 tok/sFP16 / BF16
Qwen 3 8B6.6 GBFits~32 tok/sFP16 / BF16
Qwen 3.5 9B6.4 GBFits~31 tok/sFP16 / BF16
Llama 3.1 8B6.3 GBFits~32 tok/sFP16 / BF16
DeepSeek R1 Distill Llama 8B6.3 GBFits~32 tok/sFP16 / BF16
Mistral 7B v0.35.7 GBFits~36 tok/sFP16 / BF16
Gemma 4 E4B5.7 GBFits~35 tok/sFP16 / BF16
Qwen 2.5 7B5.6 GBFits~36 tok/sFP16 / BF16
DeepSeek R1 Distill Qwen 7B5.6 GBFits~36 tok/sFP16 / BF16
Qwen 3 4B4.2 GBFits~57 tok/sFP16 / BF16
Phi-4 Mini4.1 GBFits~61 tok/sFP16 / BF16
Gemma 4 E2B4.0 GBFits~56 tok/sFP16 / BF16
Gemma 3 4B3.7 GBFits~62 tok/sFP16 / BF16
Qwen 3.5 4B3.7 GBFits~64 tok/sFP16 / BF16
Ministral 3 3B3.5 GBFits~69 tok/sFP16 / BF16
Llama 3.2 3B3.5 GBFits~72 tok/sFP16 / BF16
Llama 3.2 1B1.7 GBFits~194 tok/sFP16 / BF16
Gemma 3 1B1.6 GBFits~261 tok/sFP16 / BF16
Kimi K2.6566 GBToo small––
DeepSeek R1 671B380 GBToo small––
Llama 3.1 405B234 GBToo small––
Qwen 3.5 397B-A17B226 GBToo small––
DeepSeek V4 Flash161 GBToo small––
Qwen 3 235B-A22B135 GBToo small––
MiniMax-M2133 GBToo small––
Qwen 3.5 122B-A10B70.2 GBToo small––
Mistral Small 4 119B68.2 GBToo small––
gpt-oss-120b67.2 GBToo small––
Llama 4 Scout64.0 GBToo small––
GLM-4.5-Air62.1 GBToo small––
Qwen3-Next 80B-A3B46.3 GBToo small–Q2_K
Qwen 2.5 72B44.5 GBToo small–Q2_K
Llama 3.3 70B43.2 GBToo small–Q2_K
DeepSeek R1 Distill Llama 70B43.2 GBToo small–Q2_K

Apple M4 Pro 48GB for local AI

A Mac mini or MacBook Pro configuration that holds 32B models comfortably. About 36 GB is GPU-usable by default; raising the wired-memory limit gives more.

Setup tips: about 36 GB is GPU-usable by default, so 32B models fit at Q4_K_M with room for context. Mixture-of-experts models such as Qwen 3.6 35B-A3B run especially well on its 273 GB/s.

Our picks for the Apple M4 Pro 48GB

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.

TaskModelBest quantMax contextSpeed
Best all-rounderQwen 3.6 27BQ8_0256K~11 tok/s
CodingQwen 3.6 27BQ8_0256K~11 tok/s
Fastest repliesQwen 3.6 35B-A3BQ6_K256K~66 tok/s
Long documents (64K+)Qwen 3.8 27BQ8_0256K~11 tok/s
Images in, text outGemma 4 31BQ8_0187K~8.9 tok/s
Image generationFLUX.2 [dev]FP8––
SpecValue
Memory48 GB unified (36.0 GB GPU-usable by default)
Memory bandwidth273 GB/s
ArchitectureApple Silicon
Released2024-10
Llama 3.1 8B Q4_K_M speed~32 tok/s
Largest model that fitsMixtral 8x7B

Image and video models on the Apple M4 Pro 48GB

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

ModelFP8FP8 fitGGUF Q4Q4 fit
Stable Diffusion 1.51.8 GBFits1.7 GBFits
SDXL 1.04.1 GBFits4.1 GBFits
Stable Diffusion 3.5 Medium6.4 GBFits6.4 GBFits
Stable Diffusion 3.5 Large8.6 GBFits6.4 GBFits
FLUX.1 [dev]12.1 GBFits7.3 GBFits
FLUX.1 [schnell]12.1 GBFits7.3 GBFits
Qwen-Image20.4 GBFits12.3 GBFits
Wan 2.2 A14B (T2V / I2V)18.8 GBFits13.2 GBFits
Wan 2.2 TI2V 5B8.0 GBFits6.5 GBFits
HunyuanVideo18.3 GBFits13.1 GBFits
FLUX.1 Kontext [dev]12.3 GBFits7.5 GBFits
FLUX.2 [dev]31.0 GBFits23.6 GBFits
Z-Image Turbo6.8 GBFits4.9 GBFits
Qwen-Image-Edit 250920.5 GBFits12.4 GBFits
LTX-219.2 GBFits12.6 GBFits

Upgrade path

The next step up is the Apple M5 Pro 64GB with 64 GB and 12% more bandwidth.

Comparisons

Apple M4 Pro 48GB for LLMs: common questions

Can the Apple M4 Pro 48GB run Llama 3.3 70B?

No. It needs about 43.2 GB at Q4_K_M, far more than the 36.0 GB available.

Can the Apple M4 Pro 48GB run Qwen 3 32B?

Yes. Qwen 3 32B needs about 21.4 GB at Q4_K_M with 8K context, and the best quality that fits is Q6_K.

Can the Apple M4 Pro 48GB 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 FP16 / BF16.

How fast is the Apple M4 Pro 48GB for LLMs?

About 32 tokens per second for an 8B model at Q4_K_M, based on its 273 GB/s memory bandwidth. Larger models scale down roughly in proportion to their size.

How much of the 48 GB can the GPU use?

About 36.0 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.

Related