Archive | Working Models

Ornith 9B on a 16GB Mac mini: where a small model breaks

Jul 1, 2026

One prompt, two model sizes, the same token budget. The 9B looped on an empty tool call 39 times and never wrote the file. The 35B wrote the whole thing on the first try. Here is what broke, why the model size was the cause, and what a 9B is actually good for.

Getting Started with LM Studio: Run AI on Your Own Machine

Jun 25, 2026

LM Studio is the gentlest way into local AI. It downloads open models and runs them on your own computer, with no account, no cloud, and no per-token bill. Here is how to install it, pick a model that fits your machine, and start chatting in about ten minutes.

Run the Claude Desktop App on a Local Model with LM Studio

Jun 24, 2026

Claude Desktop has a mode that lets you point it at a model that is not Anthropic's. I wired it to LM Studio running on my own machine, fully offline, with no Claude account. It works.

I ran GLM-5.2 locally on a Mac Studio

Jun 20, 2026

I put GLM-5.2 on a Mac Studio with 512 GB of memory and ran it fully offline. The model is 368 GB, which leaves almost no room to work in. Here is the setup that loaded it, the speed it ran at, the crash that took it down twice, and an honest read on the coding output.

GGUF vs MLX on Apple Silicon

Jun 18, 2026

Everyone says the same thing: MLX is faster on Apple Silicon, and it makes sense, MLX was built for Apple's chips. I wanted to see if that holds up, so I tested it. I took Gemma 4 26B in both formats, GGUF and MLX, and ran the same tests on each, on my M3 Ultra.