Run GPT-OSS Locally on Your MacBook Pro

GPT-OSS is OpenAI's open-weights family, and it runs natively on Apple Silicon — the 20B fits a 16 GB MacBook, the 120B runs on a Studio, and both are mixture-of-experts that decode fast for their size. Here's the size that fits your Mac, and the one command that serves it.

GPT-OSS is OpenAI's open-weights family — released under a permissive license and small enough, in mixture-of-experts form, to run on your own Mac. There are two sizes: a 20B that fits a 16 GB MacBook and a 120B for Studio-class machines. Both activate only a fraction of their parameters per token, so they decode quickly, and both ship native reasoning and tool calling in OpenAI's Harmony format.

This guide is the short path: pick the right GPT-OSS 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 gpt-oss-20b                         # 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": "Explain how a bloom filter works and when to use one."}]}'

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.

Reasoning and tool calling work out of the box

serve auto-configures the Harmony chat template, tool-call parser and reasoning parser for every GPT-OSS 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. Reasoning streams separately as reasoning_content on the standard OpenAI wire, so your client sees a clean answer with the thinking demultiplexed.

That's the boring-but-load-bearing part: GPT-OSS-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 GPT-OSS.

• Cursor — chat, Ctrl-K and

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

• Codex CLI and

Aider — both speak the same OpenAI-compatible endpoint.

GPT-OSS shines at reasoning and general assistance. For the heaviest autonomous coding loops — where a model has to apply large multi-file patches in one shot — a coding-tuned family like Qwen 3.6 is a safer bet; both serve through the same endpoint, so switching is one command.

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. Because it's mixture-of-experts, the 20B decodes fast enough to feel interactive on a laptop, and the 120B is genuinely usable on a Studio.

Frequently asked questions

Can I run GPT-OSS locally on a Mac with 16 GB of RAM?

Yes — gpt-oss-20b-mxfp4-q4 is 6.0 GB on disk and fits a 16 GB Mac with room for context. For the 20B at native quality, 24 GB is comfortable (gpt-oss-20b, 10 GB). The 120B needs 48–64 GB (mxfp4) or 64 GB+ for the full weights.

What's the difference between GPT-OSS 20B and 120B?

Both are mixture-of-experts, so they decode fast for their size. The 20B fits a 16–24 GB Mac and handles everyday chat, reasoning and tool use; the 120B is meaningfully stronger at hard reasoning but needs Studio-class memory. Pick the largest that fits your RAM with headroom.

Does GPT-OSS work with Claude Code and Cursor locally?

Yes. rapid-mlx serves GPT-OSS behind an OpenAI-compatible endpoint, with the Harmony format and reasoning handled server-side, 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 GPT-OSS 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.