The Mac Mini Was a Great, Cheap Apple Computer — Until AI
The Mac Mini Was a Great, Cheap Apple Computer — Until AI
https://www.nytimes.com/wirecutter/reviews/mac-mini-ai-shortage/
Publish Date: 2026-06-01 12:32:00
Source Domain: www.nytimes.com
To see what all the fuss is about, I decided to vibe-code an AI project on a Mac mini. Vibe-coding is building software using AI without typing out code manually. I tested it out by creating a simple task manager for myself using our mini PC upgrade pick, a Mac mini with an M4 Pro processor and 48 GB of RAM. In the first test, I used local AI models on the Mac mini. Then I did the same thing using Claude Code, which relies on Anthropic’s cloud models. Local AI models have the advantage of running entirely on your machine, so you don’t have to pay continuing costs such as a monthly Claude subscription (which ranges from $20 to $100 depending on what you need it for). On the other hand, cloud models can be faster, more up-to-date, and equipped with a larger working memory, since they’re running on ultra-fast data-center servers.
Vibe-coding is not necessarily as simple or straightforward as telling a chatbot to make an app for you (although it can be). First, I had to find a model. The model is the AI, and all the other software I would use interacts with this model in different ways. Companies such as Hugging Face and Ollama store and rank the latest models, so I looked on those sites for a recent model that was large enough to fill a good portion of my 48 GB of memory and had good benchmarks compared with ChatGPT and Claude in reasoning and coding. I settled on the Qwen 3.6 35B model built by Alibaba, which takes about 30 GB to 40 GB of RAM when running and is competitive with OpenAI, Google, and Anthropic’s flagship AI models.
Then, I installed and ran Ollama, an application to run the model, and Visual Studio Code paired with an extension called Continue to add the LLM’s code to the document. I was done with setup and ready to prompt.
Source
