Summary
Can a Mac Mini replace a cloud AI subscription? Creator Alex Finn tests that question on a brand-new Mac Mini M6 with 32 GB of unified memory, following up on an earlier prediction that Apple hardware would lead local AI.
The video walks beginners through the easiest setup available. It installs Hermes Desktop, a free open-source AI agent, and uses its provider settings to run models locally. Hermes picks the best model for the machine’s memory, in this case a Qwen 3.6 35B model of roughly 21 GB, which is often compared to last year’s Claude Opus 4.5. Finn explains why Apple’s integrated memory architecture is attractive for local inference, since the whole memory pool is available to the model.
Once the model is loaded, Finn runs live prompts without pre-testing and checks response speed against timestamps. The agent also builds up memory about its user. The video then moves on to heavier workloads to see where a 32 GB machine hits its limits, and covers the strengths and weaknesses of local models, including privacy, cost and offline use.
📺 Source: Alex Finn · Published October 10, 2026
🏷️ Format: Benchmark Test
