Install Ollama and run ollama run gpt-oss:20b, a 14GB download; on a Mac it is comfortable from 24 GB of memory and tight on 16 GB. gpt-oss is OpenAI's pair of open-weight models under the Apache 2.0 license. The 20b has 21B parameters with 3.6B active at a time, and OpenAI says it runs within 16GB of memory; Grux's own catalogue estimates about 20 GB in use once the context is counted, which is why 16 GB is the edge on a Mac, where the GPU may use only about three quarters of unified memory. The 120b has 117B parameters with 5.1B active, is a 65GB download, and is built for a single 80GB GPU; on a Mac it wants 128 GB. Both carry a 128K context window. LM Studio runs them too. Grux OS (MIT) lists gpt-oss in its Local Models catalogue, finds Ollama by itself, and can use the model to act on your mail, calendar, files and shell.
brew install ollama
ollama run gpt-oss:20b # 14GB
# or, in LM Studio's command line
lms get openai/gpt-oss-20b| Your Mac's memory | gpt-oss-20b | gpt-oss-120b |
|---|---|---|
| 16 GB | Tight: it may load, with little room for context | No |
| 24 GB | Comfortable | No |
| 32 to 64 GB | Comfortable | No |
| 96 GB | Comfortable | Tight |
| 128 GB | Comfortable | Workable |
The table applies the site's rule from which LLM can my Mac run to the download sizes on Ollama's gpt-oss page, read on 3 October 2026.
| Claim | Source |
|---|---|
| 20b has 21B parameters (3.6B active) and runs within 16GB; 120b has 117B (5.1B active) for one 80GB GPU; Apache 2.0 | OpenAI, gpt-oss README |
| Ollama's downloads are 14GB and 65GB, with a 128K context window | Ollama library, gpt-oss |
lms get openai/gpt-oss-20b downloads it in LM Studio | OpenAI, gpt-oss README |
| Grux's catalogue lists gpt-oss:20b at 14 GB on disk and about 20 GB in use | Cookbook.swift |