I Unboxed the Base M6 Mac mini and Speed Ran Local AI

I bought the base M6 Mac mini with 16GB of unified memory and a 256GB SSD, then filmed the trip from opening the box to getting a useful local AI chat. The clock in the video starts with the first setup shot after unboxing and reaches about 1 hour 21 minutes by signoff. That includes a software update, recording setup, and a later retake. It is the elapsed story, not a model speed test.

The setup took longer than the unboxing

The hardware part was simple. I showed the box, added a few closeups from my phone, and used a capture card to record the first boot. Then came the ordinary setup screens and a system update. Watching a progress bar is a real part of setting up a new computer, but it does not make for much of a video, so the edit moves quickly through the waits.

I also had to get the recording working on the new machine. I installed Firefox, opened OBS Studio, and got Ollama ready. The on-screen clock keeps running across the cut-down waits and the gap before I recorded the final model attempt. It is a timeline of my setup, not a continuous shot.

My first model choice was too ambitious

I started with Gemma 4 26B through Claude Code and Ollama. I asked a tiny question and waited. The model eventually replied, which earned a little celebration in the edit, but the machine was under heavy memory pressure. With the operating system and recording apps open, this was not a useful way to start a conversation on the base configuration.

The celebration card shows elapsed setup time from after unboxing. It does not say the answer itself took that entire time. My mistake was picking a large model before finding out what felt comfortable on this particular computer.

The smaller chat was the useful first impression

For the last part of the video, I switched to Gemma 4 E4B MLX directly in Ollama's CLI. A short prompt answered promptly, and I could continue with a longer request. That was the moment the project felt like a local chat instead of a setup exercise.

I show one output rate from that session in the video, but I did not run a controlled benchmark or compare the two models under matching conditions. This first look tells me something narrower and more useful: a smaller model felt workable on my 16GB machine with other apps open. If I did the setup again, I would start small, check memory pressure, and leave room on the 256GB SSD for model files before downloading anything much larger.

Sources

• Apple's Mac mini specifications
• Ollama's Gemma 4 26B model
• Ollama's Gemma 4 E4B MLX model
• Ollama download
• Claude Code setup
• OBS Studio download
• Firefox for Mac