Apparently Qwen 4 might land this week, and it uses the same architecture as 3.8-Flash-Next (so various LLM serving tools should be able to handle it already). I'm hoping Qwen 4 includes some smaller models - the last 4B and 9B from Qwen were for 3.5.