I host a private instance of OpenWebUI connected to OpenRouter, so I can talk to new models ~immediately after they become available. I also have Qwen3.8 27b (abliterated) running in a spare PC for any queries I know will get censored by cloud LLMs.