Testing of My 24GB Mac mini use cases after 2026-07-30 Bought a M4 Mac mini. Wanted to determine what Private local AI model use cases I can rely on with it, vs where the machine breaks
Worked on:
- 2026-08-01 155438 Setting up Mac mini
- AI topic tagging agent
- Review of qwen 3 14b vs Claude Sonnet 5 writing a blog about the BVI trip
- Review of qwen 3 14b vs Claude Sonnet 5 transcript summary
Takeaways:
- Part of what you pay for with the Mac mini is the overhead from running and storing macOS and other baseline stuff. Logged stats in My 24GB Mac mini
- Realized that the context size has a big impact on the results of my tests.
qwendefault was set too low - Once I increased the context window, it was able to better handle the use cases. But the model’s generation ability was still not acceptable enough
- With more time I would test other local models outside of
qwenbut I can also do that with Rent GPUs instead of owning hardware
Result: Better to experiment with Automating sending data to AI models without owned hardware first so I can flex my hardware and model size up or down depending on the use cases. Best to 2026-08-12 return Mac mini and I can Determine need for Mac mini in the future. Will also Set up desktop as always-on machine to run frontier workflows