What LLMs Can the Apple M2 Ultra 192GB Run?

The Apple M2 Ultra 192GB has 192 GB of unified memory (about 144 GB usable by the GPU) and 800 GB/s of bandwidth. It runs 53 of 58 models fully at Q4_K_M with 8K context, up to DeepSeek V4 Flash, and generates about 93 tokens/s on an 8B model.

Apple · Apple Silicon · Released 2023-06 · Updated October 14, 2026

53 of 58 models fit

Apple M2 Ultra 192GB (144 GB GPU-usable) at Q4_K_M with 8K context. Largest that fits fully: Qwen 3 235B-A22B.

ModelNeedsStatusSpeedBest quant that fits
Qwen 3 235B-A22B135 GBFits~27 tok/sQ4_K_M
MiniMax-M2133 GBFits~50 tok/sQ4_K_M
Qwen 3.5 122B-A10B70.2 GBFits~59 tok/sQ8_0
Mistral Small 4 119B68.2 GBFits~88 tok/sFP8 (official)
gpt-oss-120b67.2 GBFits~102 tok/sMXFP4 (official)
Llama 4 Scout64.0 GBFits~38 tok/sQ8_0
GLM-4.5-Air62.1 GBFits~50 tok/sQ8_0
Qwen3-Next 80B-A3B46.3 GBFits~116 tok/sQ8_0
Qwen 2.5 72B44.5 GBFits~11 tok/sQ8_0
Llama 3.3 70B43.2 GBFits~12 tok/sFP16 / BF16
DeepSeek R1 Distill Llama 70B43.2 GBFits~12 tok/sFP16 / BF16
Mixtral 8x7B28.0 GBFits~52 tok/sFP16 / BF16
Qwen 2.5 32B21.4 GBFits~24 tok/sFP16 / BF16
Qwen 2.5 Coder 32B21.4 GBFits~24 tok/sFP16 / BF16
DeepSeek R1 Distill Qwen 32B21.4 GBFits~24 tok/sFP16 / BF16
Qwen 3 32B21.4 GBFits~24 tok/sFP16 / BF16
Qwen 3.5 35B-A3B21.0 GBFits~128 tok/sFP16 / BF16
Qwen 3.6 35B-A3B21.0 GBFits~128 tok/sFP16 / BF16
Gemma 4 31B19.8 GBFits~26 tok/sFP16 / BF16
Qwen 3 30B-A3B18.8 GBFits~106 tok/sFP16 / BF16
Qwen3-Coder 30B-A3B18.8 GBFits~106 tok/sFP16 / BF16
Gemma 3 27B17.6 GBFits~29 tok/sFP16 / BF16
Qwen 3.5 27B16.9 GBFits~30 tok/sFP16 / BF16
Qwen 3.6 27B16.9 GBFits~30 tok/sFP16 / BF16
Qwen 3.8 27B16.9 GBFits~30 tok/sFP16 / BF16
Mistral Small 3.1 24B15.6 GBFits~34 tok/sFP16 / BF16
Gemma 4 26B-A4B15.6 GBFits~125 tok/sFP16 / BF16
gpt-oss-20b12.9 GBFits~147 tok/sMXFP4 (official)
Qwen 2.5 14B10.7 GBFits~52 tok/sFP16 / BF16
DeepSeek R1 Distill Qwen 14B10.7 GBFits~52 tok/sFP16 / BF16
Phi-4 14B10.6 GBFits~52 tok/sFP16 / BF16
Qwen 3 14B10.5 GBFits~53 tok/sFP16 / BF16
Ministral 3 14B9.7 GBFits~57 tok/sFP16 / BF16
Mistral Nemo 12B9.0 GBFits~63 tok/sFP16 / BF16
Gemma 3 12B8.8 GBFits~63 tok/sFP16 / BF16
Ministral 3 8B6.7 GBFits~88 tok/sFP16 / BF16
Qwen 3 8B6.6 GBFits~90 tok/sFP16 / BF16
Qwen 3.5 9B6.4 GBFits~90 tok/sFP16 / BF16
Llama 3.1 8B6.3 GBFits~93 tok/sFP16 / BF16
DeepSeek R1 Distill Llama 8B6.3 GBFits~93 tok/sFP16 / BF16
Mistral 7B v0.35.7 GBFits~102 tok/sFP16 / BF16
Gemma 4 E4B5.7 GBFits~101 tok/sFP16 / BF16
Qwen 2.5 7B5.6 GBFits~104 tok/sFP16 / BF16
DeepSeek R1 Distill Qwen 7B5.6 GBFits~104 tok/sFP16 / BF16
Qwen 3 4B4.2 GBFits~161 tok/sFP16 / BF16
Phi-4 Mini4.1 GBFits~172 tok/sFP16 / BF16
Gemma 4 E2B4.0 GBFits~157 tok/sFP16 / BF16
Gemma 3 4B3.7 GBFits~174 tok/sFP16 / BF16
Qwen 3.5 4B3.7 GBFits~179 tok/sFP16 / BF16
Ministral 3 3B3.5 GBFits~196 tok/sFP16 / BF16
Llama 3.2 3B3.5 GBFits~203 tok/sFP16 / BF16
Llama 3.2 1B1.7 GBFits~537 tok/sFP16 / BF16
Gemma 3 1B1.6 GBFits~676 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–Q3_K_M

Apple M2 Ultra 192GB for local AI

A Mac Studio or Mac Pro configuration with 800 GB/s and about 144 GB GPU-usable by default. It runs 100–230B MoE models and 70B models at Q8_0; raising the wired-memory limit gives about 180 GB.

Our picks for the Apple M2 Ultra 192GB

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 3.5 122B-A10BQ8_0256K~59 tok/s
CodingGLM-4.5-AirQ8_0128K~50 tok/s
Fastest repliesQwen 3.5 122B-A10BQ8_0256K~59 tok/s
Long documents (64K+)Qwen 3.5 122B-A10BQ8_0256K~59 tok/s
Images in, text outGemma 4 31BFP16 / BF16256K~26 tok/s
Image generationFLUX.2 [dev]FP8––
SpecValue
Memory192 GB unified (144 GB GPU-usable by default)
Memory bandwidth800 GB/s
ArchitectureApple Silicon
Released2023-06
Llama 3.1 8B Q4_K_M speed~93 tok/s
Largest model that fitsDeepSeek V4 Flash

Image and video models on the Apple M2 Ultra 192GB

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 M3 Ultra 512GB with 512 GB and 2% more bandwidth.

Apple M2 Ultra 192GB for LLMs: common questions

Can the Apple M2 Ultra 192GB run Llama 3.3 70B?

Yes. Llama 3.3 70B needs about 43.2 GB at Q4_K_M with 8K context, and the best quality that fits is FP16 / BF16.

Can the Apple M2 Ultra 192GB 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 FP16 / BF16.

Can the Apple M2 Ultra 192GB 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 M2 Ultra 192GB for LLMs?

About 93 tokens per second for an 8B model at Q4_K_M, based on its 800 GB/s memory bandwidth. Larger models scale down roughly in proportion to their size.

How much of the 192 GB can the GPU use?

About 144 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