An Intel Mac can run a local LLM, but only through Ollama, and only on the CPU. Ollama's macOS documentation lists x86 as CPU only and requires macOS Sonoma (14) or newer, so the Intel Macs that qualify are the 2018 to 2020 MacBook Pro and MacBook Air, the 2018 Mac mini, the 2019 and 2020 iMac, the 2017 iMac Pro and the 2019 Mac Pro. LM Studio supports Apple silicon only, Jan stopped running local models on Intel Macs at v0.8.5, and Grux AI is Apple silicon only. On an Intel Mac, install Ollama and start with a small model, 2b to 4b parameters. This page quotes no speed, because we have not measured an Intel Mac.
| App | Local models on an Intel Mac | What its own documentation says |
|---|---|---|
| Ollama | Yes, CPU only | macOS 14 or newer; Apple M series (CPU and GPU) or x86 (CPU only) |
| LM Studio | No | Chip: Apple silicon (M1 to M4); Intel-based Macs are not supported |
| Jan | No, since v0.8.5 | Intel Macs unsupported since v0.8.0; since v0.8.5 local models do not run on them |
| Grux AI | No | macOS 14 or later, Apple silicon |
| GPT4All | Not tested here | macOS 12.6 or newer, best results with Apple silicon; its README does not say Intel is excluded |
Ollama is the one runner here that documents an Intel path. GPT4All's README does not exclude Intel Macs, and we have not run it on one, so treat that row as unverified.
Ollama's Mac app needs macOS Sonoma. Apple lists these Intel models as compatible with Sonoma; an older Intel Mac stops short of it and cannot install the app.
| Model | Intel years that run Sonoma |
|---|---|
| MacBook Pro | 13-inch and 15-inch 2018 and 2019, 16-inch 2019, 13-inch 2020 (Intel) |
| MacBook Air | Retina 13-inch 2018, 2019 and 2020 (Intel) |
| iMac | Retina 4K 21.5-inch 2019, Retina 5K 27-inch 2019 and 2020 |
| Mac mini | 2018 |
| iMac Pro and Mac Pro | iMac Pro 2017, Mac Pro 2019 |
Without a GPU, memory bandwidth and the CPU set the pace, so pick the smallest models. The sizes below are from the Ollama library, the same ones used on the MacBook Air page.
brew install ollama
ollama run gemma3:4b # a 3.3GB download
ollama run qwen3.5:2b # about 2.7GB to 3.1GB
How fast these answer on an Intel Mac depends on the chip and the memory, and we have not measured it, so no figure is quoted. If a reply is too slow, go one size down.
Grux AI acts on mail, calendar, files and the shell, and it needs Apple silicon. The which LLM can my Mac run page lists what each Apple silicon memory size runs, for the day you move to one.
| Claim | Source |
|---|---|
| macOS 14 or newer; x86 is CPU only | Ollama documentation, macOS |
| Apple silicon only; Intel-based Macs are not supported | LM Studio, system requirements |
| Local models do not run on Intel Macs since v0.8.5 | Jan, Mac installation |
| The Macs that run Sonoma | Apple Support, macOS Sonoma compatible computers |
| macOS 12.6 or newer, best results with Apple silicon | GPT4All README |
| Grux AI is Apple silicon only | Grux README, Requirements |