Running qwen3.6 27b through llama.cpp.

It’s about as capable as sonnet 3.5.

I use it for light scripting, but real coding is done by cloud models.

I’m also using it as the brain for my Hermes agent. It sends me digests of news, subreddits, chats that I’d like to read but don’t have time for. It does a great job researching things on the web for me, too.