By memory size

Local LLMs for a Mac
with 36GB of memory

On a Mac with 36GB of memory, models up to 18GB run comfortably alongside other apps: 23 of the 33 models measured here. The smartest is Qwen3.5-9B (84% on knowledge, 98% on math, 6.5GB). "Measured" here also counts the models measured for the series page, beyond the top page's text lineup.

Other sizes: 8GB · 16GB · 24GB · 32GB · 36GB · 48GB or more

To find your Mac's memory, choose the Apple menu → About This Mac (for example, 16 GB).

Picks by use

UseModelRule
If unsure, start heregpt-oss:20bThe smartest that replies within 5 seconds (1 more within the margin)
Smartest (if you can wait)Qwen3.5-9BThe highest share of right answers (8 more within the margin)
Replies right awaygemma3:27bThe smartest that replies within 1 second
English → Japanese translationmuse-glimmer:30bThe highest translation score (9 more within the margin)

Runs comfortably (up to 18GB)

ModelMemoryReply timeKnowledgeMathEN→JA
Qwen3.5-9BAlibaba6.5GBabout 40.3 s84%98%—
gemma-4-12B-itGoogle8.1GBabout 7.5 s85%97%35 pts
gpt-oss:20bOpenAI12.8GBabout 1.8 s83%98%33 pts
muse-glimmer:30bMeta17.6GBabout 15 s83%98%36 pts
Ornith-1.0-9Bornith-ai6.2GBabout 3.8 s85%94%—
Qwen3-14BAlibaba10.6GBabout 8 s82%96%34 pts
Qwen3.5-4BAlibaba3.5GBabout 29.7 s81%96%—
LFM2.5-24B-A2BLiquid AI14.9GBabout 7.1 s81%96%28 pts
NVIDIA-Nemotron-Nano-12B-v2NVIDIA8.1GBabout 6.8 s80%94%31 pts
Ornith-1.5-9Bornith-ai6.4GBabout 2.5 s77%94%33 pts
gemma-4-E4B-itGoogle5.4GBabout 2.6 s79%92%32 pts
NVIDIA-Nemotron-3.5-Bolt-E4B-BF16NVIDIA6.5GBabout 2.6 s76%93%34 pts
gemma3:27bGoogle17.5GBabout 0.4 s76%91%33 pts
Qwen3-4BAlibaba3.9GBabout 3.1 s72%93%29 pts
MiniCPM5-2BOpenBMB2GBabout 1.5 s73%90%23 pts
DeepSeek-R1-0528-Qwen3-8BDeepSeek6.4GBabout 9.4 s72%89%23 pts
LFM2.5-2.6BLiquid AI2GBabout 2.3 s73%79%29 pts
NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA3.3GBabout 1.9 s72%80%24 pts
LFM2.5-8B-A1BLiquid AI5.4GBabout 1.7 s71%78%29 pts
GLM-4-9B-0414Z.ai6.7GBabout 0.1 s72%77%32 pts
LFM2-24B-A2BLiquid AI14.8GBabout 0.1 s53%66%32 pts
LFM2.5-1.2B-JP-202606Liquid AI1GBabout 0.2 s51%63%33 pts
DeepSeek-R1-Distill-Qwen-1.5BDeepSeek1.5GBabout 1.7 s23%21%3 pts

Runs with heavy apps closed

More than half the memory, but up to 23.4GB should run. Close heavy apps other than a browser first.

ModelMemoryReply timeKnowledgeMathEN→JA
Ornith-1.0-35Bornith-ai21.6GBabout 9.1 s89%100%32 pts
qwen3.8:27bAlibaba18.2GBabout 7.7 s87%100%31 pts
qwen3.6:35b-a3bAlibaba22.3GBabout 8.5 s87%98%29 pts
gemma4:26bGoogle18.7GBabout 3.2 s85%98%34 pts
Ornith-1.5-35B-A3Bornith-ai21.8GBabout 2.5 s83%97%32 pts
DeepSeek-R1-Distill-Qwen-32BDeepSeek22.4GBabout 17.9 s81%94%29 pts
glm-4.7-flash:latestZ.ai19.4GBabout 7.5 s76%95%32 pts
qwen3-coder:30bAlibaba19.4GBabout 0.1 s69%89%31 pts
Hy-Memory-30B-A3BTencent19.2GBabout 0.1 s66%88%31 pts
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16NVIDIA22.7GBabout 3.7 s80%60%28 pts

Image and video models (guide)

Models whose measured peak memory is up to 23.4GB. Close other apps first to be safe.

Apple list price, tax included (guide)

The tax-included list price is unconfirmed. The buy-page terms forbid automated fetching, so this site did not retrieve yen amounts.

MemoryApple list price, tax included (guide)Difference from the size below
8GBUnconfirmed—
16GBUnconfirmedUnconfirmed
24GBUnconfirmedUnconfirmed
32GBUnconfirmedUnconfirmed
36GBUnconfirmedUnconfirmed
48GB or moreUnconfirmedUnconfirmed

Source: Apple’s official buy pages (www.apple.com/jp/shop/buy-mac/mac-mini / www.apple.com/jp/shop/buy-mac/mac-studio). Checked 2026-10-24. No outbound link.

About this guide

  • Memory was measured on a Mac Studio (M5 Max, 36GB). Memory use is about the same on any Apple silicon Mac, so it carries over.
  • Reply times are from this Mac. Slower chips (a MacBook Air, say) reply more slowly.
  • "Comfortable" means the model takes at most half the memory; "with heavy apps closed" up to 65%. The picks follow the same rules as the text page, among the models that fit.
  • What "9B", "A3B" or "it" in the names mean: how to read a model name.

Task recommendations and capabilities · Measurements by model