Benchmarks

A custom 26-test benchmark across seven categories, run on my home AI box to see which local LLM actually wins.