Run Llama Locally on Your MacBook Pro

Llama is the model that made local LLMs mainstream — and the 3.x line is small enough to run on any Mac, from a 0.5 GB 1B on a base Air to an 8B that's comfortable on 16 GB. Here's the size that fits your Mac, and the one command that serves it.

Llama is the family that made running LLMs on your own hardware a normal thing to do. Meta's open weights are permissively licensed, widely supported, and — crucially for a Mac — the Llama 3.x line stays small. The biggest one here is an 8B; the smallest is a 0.5 GB 1B that runs on a base MacBook Air with room to spare. All of it runs natively on Apple Silicon.

This guide is the short path: pick the right Llama for your Mac's RAM, serve it with one command, and point your tools at it.

The 60-second version

curl -fsSL https://rapidmlx.com/install.sh | bash   # or: brew install rapid-mlx
rapid-mlx serve llama-3.1-8b-4bit                   # pick your size below

Then talk to it — it's a standard OpenAI-compatible server on http://localhost:8000:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "default", "messages": [{"role": "user", "content": "Summarize the plot of Moby-Dick in three sentences."}]}'

First run downloads the weights; after that, startup is seconds. No account, no API key, nothing leaves your machine.

Which size fits your Mac

Sizes below are what the weights occupy on disk (and roughly in unified memory once loaded). Leave headroom for your context window and the rest of macOS.

Not sure where you land? The hardware tier guide maps every Mac from an 8 GB Air to a 512 GB Studio to a recommendation, and models.rapidmlx.com lets you filter the whole catalog by your RAM.

Tool calling works out of the box

serve auto-configures the right chat template and tool-call parser for every Llama alias, and grammar-constrained tool calling is on by default — so function calls come back as valid JSON, not as a paragraph that almost parses. That's the boring-but-load-bearing part: Llama-on-your-Mac slots into tools that expect OpenAI semantics without any glue code.

Wire it into your editor or agent

The server speaks the OpenAI protocol, so anything with a "custom base URL" box can use it:

• Claude Code — run the CLI

agent fully local against your Llama.

• Cursor — chat, Ctrl-K and

composer against http://localhost:8000/v1.

• Codex CLI and

Aider — both speak the same OpenAI-compatible endpoint.

Llama's sweet spot is fast, general-purpose chat and summarization that fits in very little memory. For long autonomous coding loops, prefer a larger model — but for a quick local assistant that runs anywhere, the 8B is hard to beat.

How fast is it?

We publish real numbers rather than vibes: our five-model MacBook Pro benchmark puts local models through coding, reasoning, tool-calling and memory tests, and the M3 Ultra benchmark covers the desktop end. Being small, the Llama 3.x models decode very fast on Apple Silicon — the 1B and 3B are near-instant, and the 8B is comfortably interactive on a laptop.

Frequently asked questions

Can I run Llama locally on a Mac with 8 GB of RAM?

Yes — llama3-3b-4bit (Llama 3.2 3B) is 1.5 GB on disk and runs comfortably on an 8 GB MacBook Air, and llama3-1b-4bit is just 0.5 GB for the lightest tasks. For a stronger general-purpose model, 16 GB fits the 8B (llama-3.1-8b-4bit) with room for long context.

Which Llama should I use for coding on a Mac?

The Llama 3.x line here tops out at 8B, which handles everyday code questions and small edits well. For heavier, multi-step agent coding, point your tool at a larger family such as Qwen 3.6 or Gemma 4 — rapid-mlx serves them behind the same OpenAI-compatible endpoint, so switching is one command.

Does Llama work with Claude Code and Cursor locally?

Yes. rapid-mlx serves Llama behind an OpenAI-compatible endpoint, so any tool with a custom base URL box — Claude Code, Cursor, Codex, Aider — can point at http://localhost:8000/v1 and use it with no bridge and no plugin.

Is running Llama locally actually free and private?

The weights are open, rapid-mlx is Apache 2.0, and inference happens entirely on your Mac — no account, no per-token billing, and your prompts and code never leave the machine. The only costs are disk space for the weights and the electricity your Mac was already using.