Will It Run: check if an AI model fits your Mac
- Will It Run is a free helper your AI assistant can call:
https://mcp.outlier.host/mcp. It is read-only and needs no account. - Ask Claude or ChatGPT things like “Will Qwen3.6-27B run on my 16 GB MacBook Air M2?” and it answers from a dated table, not a guess.
- Memory fit comes from our own measurements. Speed is measured on one Mac (an M1 Ultra) and estimated for the others, and every answer says which.
Add it to your assistant
Claude: Settings › Connectors › Add custom connector, name it Will It Run, and paste https://mcp.outlier.host/mcp. Until the directory listing is live, this is the way in.
ChatGPT: once it is listed in ChatGPT’s app directory, add it from there. Until then it is Claude only.
Anything else that speaks MCP: point it at the same address. It uses the Streamable HTTP transport and has no login.
What it answers, with real examples
Each of these was asked on the live server on 7 November 2026; the answers are the server’s own words, shortened.
| You ask | It answers |
|---|---|
| Will Qwen3.6-27B run on my 16 GB MacBook Air M2? | Does not fit: it needs 24 GB, so it is 8 GB short. The download is 15.13 GB. |
| What can I run on a 16 GB MacBook Air M2? | Qwen3.5-4B fits with 10 GB to spare, Qwen3.5-9B with 4 GB, and Gemma-4-26B-a4b-it is tight at 0 GB, each with a speed range. |
| Compare a 24 GB MacBook Air 13 M4 with a 48 GB Mac mini M4 Pro for local AI. | On the Air both 27B models are tight and the 397B does not fit; on the mini everything but the 397B fits. |
| Which Mac runs Qwen3.6-27B at 10 tokens a second or more? | Only the Ultra Mac Studios in the table. |
The six tools
- check_fit: does one model fit one Mac and memory size, and how much room is left.
- models_that_fit: every model that fits a Mac, for chat, coding, reading images or anything.
- computer_specs: a Mac’s chip, cores, memory options and memory bandwidth.
- speed_estimate: roughly how many tokens a second a model runs at on a Mac.
- compare_macs: two Macs side by side.
- pick_mac: which Macs and memory sizes run the models you want.
What is in the table
- Macs: 38 Apple-silicon Macs from Apple’s technical specifications, from the 2020 M1 to the Macs Apple sells today: the MacBook Neo and the M5 and M6 Macs, up to the Mac Studio M5 Ultra (Apple’s pages, read 7 November 2026).
- Models: six open models: Qwen3.5-4B, Qwen3.5-9B, Gemma-4-26B-a4b-it, Qwen3.6-27B, Qwen3.8-27B and Qwen3.5-397B-A17B, each with its minimum memory and download size from our own tests.
- Speed: measured on a Mac Studio M1 Ultra with 64 GB (800 GB/s). For other Macs it shows a range labelled as an estimate: the low end scales that measurement by memory bandwidth, and the high end uses the chip’s speed in llama.cpp’s public Apple-silicon benchmark, which shows smaller chips running faster than bandwidth alone suggests. A speed floor in pick_mac uses the low end. The 397B runs from the SSD, so it gets no estimate at all.
- Context: memory figures are measured at a 32K context. Longer contexts need more memory.
Limits
- Macs only for now, and only the Macs and models in the table. An unknown name gets the closest matches, never a made-up answer.
- No prices. We keep no dated price list, so it never quotes one.
- One measured machine for speed. More are coming.
Privacy
Your assistant sends only the tool’s inputs (a Mac name, a memory size, model names, a task). The service keeps nothing and contacts no one else. The full section is in our privacy policy.
Support
Questions, a wrong answer, or a Mac or model you want added: email matt@outlier.host. A person reads every message.
Questions
Is Will It Run free?
Yes. It needs no account, no key and no plan. It reads a fixed table and changes nothing.
Where do the numbers come from?
Apple's technical specifications for the Macs, and our own measurements for the models: each model's minimum memory, and its speed measured on a Mac Studio M1 Ultra with 64 GB. Speeds on other Macs are estimates, and every answer says which it is.
Does it see my conversation?
No. Your assistant sends only the tool's inputs: a Mac name, a memory size, model names, a task and, if you asked, a minimum speed.
Run these models on your own Mac
Outlier runs the same open models offline, sized to your Mac’s memory. No account, no token bill.
Get Outlier