What LLMs Can the Apple M5 32GB Run?

The Apple M5 32GB has 32 GB of unified memory (about 21.3 GB usable by the GPU) and 153 GB/s of bandwidth. It runs 37 of 58 models fully at Q4_K_M with 8K context, up to Qwen 3.5 35B-A3B, and generates about 18 tokens/s on an 8B model.

Apple · Apple Silicon · Released 2025-10 · Updated October 14, 2026

37 of 58 models fit

Apple M5 32GB (21.3 GB GPU-usable) at Q4_K_M with 8K context. Largest that fits fully: Qwen 3.5 35B-A3B.

ModelNeedsStatusSpeedBest quant that fits
Gemma 4 31B19.8 GBFits~5.0 tok/sQ4_K_M
Qwen 3 30B-A3B18.8 GBFits~35 tok/sQ4_K_M
Qwen3-Coder 30B-A3B18.8 GBFits~35 tok/sQ4_K_M
Gemma 3 27B17.6 GBFits~5.7 tok/sQ4_K_M
Qwen 3.5 27B16.9 GBFits~5.9 tok/sQ5_K_M
Qwen 3.6 27B16.9 GBFits~5.9 tok/sQ5_K_M
Qwen 3.8 27B16.9 GBFits~5.9 tok/sQ5_K_M
Mistral Small 3.1 24B15.6 GBFits~6.5 tok/sQ5_K_M
Gemma 4 26B-A4B15.6 GBFits~34 tok/sQ5_K_M
gpt-oss-20b12.9 GBFits~39 tok/sMXFP4 (official)
Qwen 2.5 14B10.7 GBFits~10 tok/sQ8_0
DeepSeek R1 Distill Qwen 14B10.7 GBFits~10 tok/sQ8_0
Phi-4 14B10.6 GBFits~10 tok/sQ8_0
Qwen 3 14B10.5 GBFits~10 tok/sQ8_0
Ministral 3 14B9.7 GBFits~11 tok/sQ8_0
Mistral Nemo 12B9.0 GBFits~12 tok/sQ8_0
Gemma 3 12B8.8 GBFits~12 tok/sQ8_0
Ministral 3 8B6.7 GBFits~17 tok/sFP16 / BF16
Qwen 3 8B6.6 GBFits~18 tok/sFP16 / BF16
Qwen 3.5 9B6.4 GBFits~18 tok/sFP16 / BF16
Llama 3.1 8B6.3 GBFits~18 tok/sFP16 / BF16
DeepSeek R1 Distill Llama 8B6.3 GBFits~18 tok/sFP16 / BF16
Mistral 7B v0.35.7 GBFits~20 tok/sFP16 / BF16
Gemma 4 E4B5.7 GBFits~20 tok/sFP16 / BF16
Qwen 2.5 7B5.6 GBFits~20 tok/sFP16 / BF16
DeepSeek R1 Distill Qwen 7B5.6 GBFits~20 tok/sFP16 / BF16
Qwen 3 4B4.2 GBFits~32 tok/sFP16 / BF16
Phi-4 Mini4.1 GBFits~34 tok/sFP16 / BF16
Gemma 4 E2B4.0 GBFits~31 tok/sFP16 / BF16
Gemma 3 4B3.7 GBFits~35 tok/sFP16 / BF16
Qwen 3.5 4B3.7 GBFits~36 tok/sFP16 / BF16
Ministral 3 3B3.5 GBFits~39 tok/sFP16 / BF16
Llama 3.2 3B3.5 GBFits~41 tok/sFP16 / BF16
Llama 3.2 1B1.7 GBFits~110 tok/sFP16 / BF16
Gemma 3 1B1.6 GBFits~151 tok/sFP16 / BF16
Qwen 3.5 35B-A3B21.0 GBTight fit~42 tok/sIQ4_XS
Qwen 3.6 35B-A3B21.0 GBTight fit~42 tok/sIQ4_XS
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––
Qwen 2.5 72B44.5 GBToo small––
Llama 3.3 70B43.2 GBToo small––
DeepSeek R1 Distill Llama 70B43.2 GBToo small––
Mixtral 8x7B28.0 GBToo small–Q2_K
Qwen 2.5 32B21.4 GBToo small–IQ4_XS
Qwen 2.5 Coder 32B21.4 GBToo small–IQ4_XS
DeepSeek R1 Distill Qwen 32B21.4 GBToo small–IQ4_XS
Qwen 3 32B21.4 GBToo small–IQ4_XS

Apple M5 32GB for local AI

The base M5 adds neural accelerators to each GPU core, which speeds up prompt processing, but 153 GB/s of bandwidth still limits token generation. About 21 GB is GPU-usable by default.

Our picks for the Apple M5 32GB

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 27BQ5_K_M58K~5.9 tok/s
CodingQwen 3.6 27BQ5_K_M58K~5.9 tok/s
Fastest repliesgpt-oss-20bMXFP4 (official)128K~37 tok/s
Long documents (64K+)Gemma 4 26B-A4BQ5_K_M233K~34 tok/s
Images in, text outGemma 4 31BQ4_K_M10K~5.0 tok/s
Image generationQwen-ImageGGUF Q6_K––
SpecValue
Memory32 GB unified (21.3 GB GPU-usable by default)
Memory bandwidth153 GB/s
ArchitectureApple Silicon
Released2025-10
Llama 3.1 8B Q4_K_M speed~18 tok/s
Largest model that fitsQwen 3.5 35B-A3B

Image and video models on the Apple M5 32GB

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 GBTight fit12.3 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 GBToo small23.6 GBToo small
Z-Image Turbo6.8 GBFits4.9 GBFits
Qwen-Image-Edit 250920.5 GBTight fit12.4 GBFits
LTX-219.2 GBFits12.6 GBFits

Upgrade path

The next step up is the Apple M4 Max 36GB with 36 GB and 168% more bandwidth.

Apple M5 32GB for LLMs: common questions

Can the Apple M5 32GB run Llama 3.3 70B?

No. It needs about 43.2 GB at Q4_K_M, far more than the 21.3 GB available.

Can the Apple M5 32GB run Qwen 3 32B?

No. It needs about 21.4 GB at Q4_K_M, far more than the 21.3 GB available.

Can the Apple M5 32GB 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 Q8_0.

How fast is the Apple M5 32GB for LLMs?

About 18 tokens per second for an 8B model at Q4_K_M, based on its 153 GB/s memory bandwidth. Larger models scale down roughly in proportion to their size.

How much of the 32 GB can the GPU use?

About 21.3 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