Which local LLM can a MacBook Air run?

A MacBook Air runs local LLMs well, and its memory decides which ones. The current Air ships with 16 GB, configurable to 24 GB or 32 GB; the M1 Air shipped with 8 GB, configurable to 16 GB. On 8 GB, run a small model such as Gemma 3 4b (3.3GB) or Qwen 3.5 2b (2.7GB to 3.1GB). On 16 GB, Qwen 3.5 9b (6.6GB to 7.6GB) or Gemma 4 12b (7.7GB to 8.0GB) fit with room for context. On 24 GB those run comfortably, and gpt-oss 20b (14GB) becomes possible. On 32 GB, gpt-oss 20b is comfortable and Qwen 3.8 27b (18GB) is tight. Install Ollama, run one command, and the model answers on the Mac with nothing sent anywhere. Grux OS (MIT) finds Ollama by itself and can use the same model to act on your mail, calendar, files and shell.

Grux OS 3.0.0 · last checked 2026-09-30 · generated from the shipping release

By memory

MacBook Air memoryModels that fitDownload
8 GBGemma 3 4b, Qwen 3.5 2b3.3GB, 2.7GB to 3.1GB
16 GBQwen 3.5 9b, Gemma 4 12b6.6GB to 7.6GB, 7.7GB to 8.0GB
24 GBThe 16 GB models comfortably; gpt-oss 20b is tight14GB
32 GBgpt-oss 20b comfortably; Qwen 3.8 27b is tight14GB, 18GB

The table applies the rule on the which LLM can my Mac run page: the GPU may use about three quarters of unified memory, and the model file should stay under about two thirds of that, leaving room for context. Memory decides which model fits; a newer chip runs the same model faster.

Start one

brew install ollama
ollama run gemma3:4b       # 8 GB Air
ollama run qwen3.5         # 16 GB Air, the 9b model
ollama run gpt-oss:20b     # 24 GB or 32 GB Air

Each model has its own page here with every size: Qwen, Gemma and gpt-oss.

Where to check this

ClaimSource
The current MacBook Air: 16GB, configurable to 24GB or 32GBApple, MacBook Air tech specs
The M1 MacBook Air: 8GB, configurable to 16GBApple Support, MacBook Air M1 tech specs
Qwen 3.5 sizes; Qwen 3.8 27b at 18GBOllama library, qwen3.5 and qwen3.8
Gemma 4 and Gemma 3 sizesOllama library, gemma4 and gemma3
gpt-oss 20b is a 14GB downloadOllama library, gpt-oss
Grux runs on Ollama running locally, with no keyGrux README, Requirements

Questions

Can an M1 MacBook Air with 8 GB run a local LLM?
Yes, a small one. Gemma 3 4b at 3.3GB or Qwen 3.5 2b fits; larger models will not leave room for context.
Is 16 GB enough for a local LLM on a MacBook Air?
Yes. Qwen 3.5 9b or Gemma 4 12b fits with room for context, and both read images as well as text.
Which MacBook Air is best for local LLMs?
The one with the most memory. Memory decides which model fits, so 24 GB or 32 GB opens up gpt-oss 20b; a newer chip runs the same model faster.
Download Grux 3.0.0 Read the source

Free, MIT licensed. macOS 14 or later, Apple silicon. 23.7 MB, signed and notarized by Apple. No account, no server, no subscription.