How to run AI on a Mac with 8GB of RAM

Plenty of Macs have 8GB, M1 MacBook Airs among them. If yours does, the question is fair: is that enough for local AI? It is, for small models. This page shows which ones fit, which do not, and what to do first. The free download button is at the bottom.

The RAM math: what's actually free

On an Apple Silicon Mac, the CPU, GPU and Neural Engine share one pool of unified memory. That helps AI speed, because the model does not get copied across a PCIe bus. The catch: macOS takes its share before your model loads.

With a typical app load (browser, a few tabs, maybe a notes app), macOS and your running apps use roughly 2–4GB of the 8GB. That leaves 4 to 6GB for a model. Some models fit easily in that space. Others do not fit at all.

The number that matters for local AI is wired memory, the part the GPU holds and will not swap out. At 4-bit quantization (the standard compression format for running models locally), wired usage is well below the raw parameter count. That is why a 7B model does not need 7GB.

Which model sizes fit on 8GB

Here is how common model sizes map to wired memory at 4-bit quantization, and whether each is practical on 8GB. The wired figures are approximate.

The cutoff is real. If a model's wired footprint is bigger than your free memory, macOS swaps: it writes model weights to the SSD and reads them back on demand. That slows inference to roughly 1–3 tok/s and heats the Mac up, with the fan spinning on any Mac that has one. It does not crash, but it is not usable. The fix is to stay inside your memory budget.

What Nano is actually good for

Nano (4B) is a good fit for everyday tasks:

• Chat and Q&A. Answering questions, explaining concepts, working through ideas.
• Writing and editing. Drafting emails, rewriting paragraphs, tightening copy.
• Summarization. Paste in a long document and get the key points, with no data leaving your machine.

Nano's limits show on harder work: multi-step coding, long reasoning chains, and tasks that need a lot of context. On Outlier's own MMLU run, Nano scored 52.0% (104 of 200), Lite 77.0% (154 of 200) and Core, the 27B model, 89.5% (179 of 200). Core needs 24GB, so that is a different Mac. For daily writing, research and quick coding help, start with Nano.

Pushing it with 7B: what to close, what to expect

A 7B model at 4-bit quantization uses around 4.5GB wired. That fits on an 8GB Mac, barely. Whether it runs well depends on what else you have open. Outlier does not ship a 7B tier, so you would import one you downloaded yourself.

Before loading a 7B model on 8GB, quit the apps eating memory. Quit them, do not minimize them:

• Close Safari or Chrome, or at minimum close all but a handful of tabs
• Quit Slack, Teams, or any Electron-based app
• Quit mail clients, music apps, and anything running in the background

With a clean slate, you should have 5–6GB free. That is enough for a 7B model to load without hitting swap. Expect 10–15 tok/s, depending on context length. That is slower than Nano, but it runs, and the quality step-up over Nano is real for tasks that need it.

If macOS starts swapping, you will see it: the Mac heats up and responses arrive one token at a time. Step back down to Nano.

When you'll want more RAM

8GB is a workable machine for local AI. It has limits. More memory earns its cost when:

• You want to run a 7B model without quitting your other apps first, or you switch between AI and other memory-hungry apps all day.
• You want to try 13B or larger models, which swap on 8GB.
• You do long-context work (large codebases, multi-document summarization), where memory pressure is higher.

If you chat, write and ask questions with Nano, 8GB is enough. If you want Lite (9B), you need 12GB or more. If you want 7B models without quitting other apps first, 16GB removes the friction. The 16GB page covers what that step buys you.

Step by step: running AI on your 8GB Mac with Outlier

1. Download Outlier with the free button below, or from outlier.host. It is 290 MB and a signed Mac app. No terminal, no Python, no account.
2. Open Outlier. On first launch, it downloads the Nano model. The file is 2.37GB and downloads once.
3. Start chatting. Try a summary, a draft, a question. Speed depends on your Mac.
4. Stay on Nano. Lite (9B) is free too, but Outlier lists 12GB as its minimum. On an 8GB Mac it swaps instead of running well.
5. To try 7B, quit your heavy background apps first (see above), then import the model as the import guide describes. Watch Activity Monitor → Memory to check you are not swapping.

Your prompts run on your chip and stay on your disk. Nothing you type leaves your Mac unless you switch on web search. Once the Nano download finishes, you can turn Wi-Fi off and it still answers.