CanMyPCRunAILocal AI compatibility

What AI models can a Mac with 32GB of memory run?

On an Apple Silicon Mac with 32GB of unified memory, 185 of 224 catalog builds have good/excellent reference fits at 4K context, and 7 more run with trade-offs. The memory is shared with macOS and your other apps, so the headroom is smaller than the number suggests.

185

run well

7

with trade-offs

224

builds evaluated

Check my actual Mac

Unified memory is not VRAM

A PC graphics card has its own memory, separate from system RAM. A Mac with Apple Silicon has one pool that the CPU and GPU share. That is why the answer differs from a 32GB graphics card, and why a model never gets the whole 32GB: macOS and open apps use some of it. These pages use a reference scenario. They do not know which M-series chip you have, and Apple sells several chips with the same memory size at very different speeds, so no speed is claimed here.

Four builds to compare by task

Editorial examples chosen for their documented tasks, all Q4_K_M at 4,096 tokens. Check the verdict before downloading. These are estimates, not benchmark rankings.

Short conversations and summaries

llama3.2:3b

A small instruction-tuned model for trying local chat with more memory headroom. Review summaries against the source text.

Task and build source · Catalog checked 2026-09-13

Excellent fit

Est. memory 2.9 GB · high confidence

Model fits within the unified memory GPU allocation.

ollama run llama3.2:3b

General instruction following

llama3.1:8b

An 8B instruction-tuned alternative to compare on your own prompts. It needs more memory than the 3B example and can still give incorrect answers.

Task and build source · Catalog checked 2026-09-13

Excellent fit

Est. memory 5.8 GB · high confidence

Model fits within the unified memory GPU allocation.

ollama run llama3.1:8b

Code explanation and editing

qwen2.5-coder:7b

A code-focused instruct build. Review generated changes and run your tests; a memory fit is not evidence that its code is correct.

Task and build source · Catalog checked 2026-09-13

Excellent fit

Est. memory 5.3 GB · high confidence

Model fits within the unified memory GPU allocation.

ollama run qwen2.5-coder:7b

A larger general-purpose comparison

qwen2.5:14b

A 14B instruct build for comparing answers when memory allows. Larger size does not establish better results for every task.

Task and build source · Catalog checked 2026-09-13

Excellent fit

Est. memory 10.2 GB · high confidence

Model fits within the unified memory GPU allocation.

ollama run qwen2.5:14b

After opening Ollama's prompt, enter /set parameter num_ctx 4096 to match this calculation, within the model's supported maximum. Install a current Ollama release that supports the build. Setup and troubleshooting.

Models that fit — largest memory first — on a 32GB Mac

Models that fit in 32GB of unified memory
ModelBuildMemory neededVerdict
qwen3.8qwen3.8:27bQ4_K_M~22.8 GBGood fit
muse-glimmermuse-glimmer:30bQ4_K_M~22.7 GBGood fit
mistral-small3.1mistral-small3.1:24bQ4_K_M~21.0 GBGood fit
devstral-small-2devstral-small-2:24bQ4_K_M~20.6 GBGood fit
mistral-small3.2mistral-small3.2:24bQ4_K_M~20.6 GBGood fit
lfm2lfm2:24bQ4_K_M~19.6 GBGood fit
devstraldevstral:24bQ4_K_M~19.5 GBGood fit
magistralmagistral:24bQ4_K_M~19.5 GBGood fit
gpt-ossgpt-oss:20bMXFP4~18.8 GBExcellent fit
gpt-oss-safeguardgpt-oss-safeguard:20bMXFP4~18.8 GBExcellent fit
mistral-smallmistral-small:22b-instruct-2409-q4_K_MQ4_K_M~18.2 GBExcellent fit
codestralcodestral:22b-v0.1-q4_K_MQ4_K_M~18.2 GBExcellent fit
solar-prosolar-pro:22bQ4_K_M~18.1 GBExcellent fit
phi4-reasoningphi4-reasoning:14bQ4_K_M~15.2 GBExcellent fit
deepseek-coder-v2deepseek-coder-v2:16b-lite-base-q4_K_MQ4_K_M~14.2 GBExcellent fit
deepseek-v2deepseek-v2:16b-lite-chat-q4_K_MQ4_K_M~14.2 GBExcellent fit
clef-flashclef-flash:9bQ8_0~13.7 GBExcellent fit
phi4phi4:14bQ4_K_M~12.4 GBExcellent fit
everythinglmeverythinglm:13b-16k-q4_K_MQ4_K_M~10.9 GBExcellent fit
nexusravennexusraven:13b-q4_K_MQ4_K_M~10.9 GBExcellent fit
open-orca-platypus2open-orca-platypus2:13b-q4_K_MQ4_K_M~10.9 GBExcellent fit
codeupcodeup:13b-llama2-chat-q4_K_MQ4_K_M~10.9 GBExcellent fit
wizard-vicunawizard-vicuna:13b-q4_K_MQ4_K_M~10.9 GBExcellent fit
wizardlm-uncensoredwizardlm-uncensored:13b-llama2-q4_K_MQ4_K_M~10.9 GBExcellent fit
llama3.2-visionllama3.2-vision:11bQ4_K_M~10.8 GBExcellent fit

Showing the 25 largest of 185 models that run well.

How this was calculated

Each model is evaluated at 4K context against 32GB of shared memory on a reference Apple Silicon Mac running the Metal backend, counting weights, KV cache, compute buffers and runtime overhead. The engine reserves memory for macOS and estimates the share available to the GPU. That share is a heuristic. Available memory and current memory pressure are unknown; check Activity Monitor during a real run. Ollama requires macOS 14 or later. See the 16 GB Mac walkthrough for a worked example, or the scan to measure your own Mac.

32GB Mac and local AI — frequently asked questions

Can a Mac with 32GB of memory run local AI models?
185 of the 224 builds evaluated have good/excellent reference fits with 32GB of unified memory at 4K context, and 7 more run with trade-offs. The heaviest that fits is qwen3.8:27b, needing about 22.8 GB.
Is unified memory the same as VRAM?
No. On Apple Silicon the 32GB is shared by the CPU, the GPU, macOS and every open app, so a model never gets all of it. The engine reserves room for the system and estimates the share the GPU can use; that share is a heuristic, not a measurement of your Mac.
Does this apply to an Intel Mac?
No. Ollama uses the GPU only on Apple M-series Macs. An Intel Mac runs models on the CPU alone, which is much slower, so do not apply these results to it.

Other Mac memory sizes

8GB16GB24GB48GB64GB96GB128GB