Day 5: Local LLM Exceeds Expectation

The local llm homelab is way surpassing what I had expected. Qwen 3.8-27b is much more capable than I imagined. It's roughly on par with sonnet imo, and i'm pretty trusting of it's code output for small and targeted tasks.

Also being able to fit 128k context into one 32GB card has allowed me to run multiple agents in parallel - something i was not expecting based on some initial benchmarks I read.

All in all, the open weight model is capable, small, and gets the job done without much tinkering. Happy with that!