Install Ollama and run ollama run deepseek-r1, which pulls the 8b model, about 5.2GB, and opens a chat in the terminal. Pick the size by your Mac's memory: the 1.5b model (1.1GB) suits 8 GB, the 7b or 8b (4.7GB to 5.2GB) suits 16 GB, the 14b (9.0GB) wants 24 GB, the 32b (20GB) is comfortable from 64 GB, and the 70b (43GB) needs 96 GB or more. The full model, 671b at 404GB, is beyond almost every Mac. The smaller sizes are distilled: DeepSeek's reasoning trained into Qwen and Llama models, so they are related to the full model but not the same. The weights are MIT licensed and run with a 128K context window. Once it runs, point any app that speaks Ollama at it, or install Grux OS (MIT), which finds Ollama by itself and can use the model to act on your mail, calendar, files and shell.
brew install ollama
ollama run deepseek-r1 # the 8b model, about 5.2GB
ollama run deepseek-r1:14b # 9.0GB, for 24 GB of memory or more| Your Mac's memory | DeepSeek R1 size | Download |
|---|---|---|
| 8 GB | 1.5b (the 7b is tight) | 1.1GB |
| 16 GB | 7b or 8b | 4.7GB to 5.2GB |
| 24 GB | 14b | 9.0GB |
| 32 GB | 14b comfortably; 32b is tight | 9.0GB, 20GB |
| 64 GB | 32b | 20GB |
| 96 GB or more | 70b | 43GB |
The table applies the rule on the which LLM can my Mac run page: the GPU may use about three quarters of unified memory, and the model file should stay under about two thirds of that, leaving room for context. Sizes are from Ollama's DeepSeek R1 page, read on 3 October 2026.
Only the 671b model is DeepSeek R1 itself. The 1.5b, 7b, 14b and 32b are Qwen models and the 70b a Llama model, fine-tuned on reasoning data DeepSeek R1 generated; the default 8b is DeepSeek-R1-0528-Qwen3-8B. They reason the same way at a smaller size, but expect less than the full model.
| Claim | Source |
|---|---|
| Sizes from 1.5b (1.1GB) to 671b (404GB), MIT weights, 128K context | Ollama library, deepseek-r1 |
| The smaller models are distilled into Qwen and Llama | Ollama library, deepseek-r1 |
Ollama installs on macOS and ollama run chats with a model | Ollama README |
| Grux runs on Ollama running locally, with no key | Grux README, Requirements |