Open WebUI Breaks Qwen 3.6 Thinking

This looks more like a client-side compatibility bug than a model issue: the backend can emit reasoning, but Open WebUI may be serializing or replaying it in the wrong shape. Open WebUI’s docs say it can only preserve reasoning that the model actually returns, and the GitHub discussion suggests `reasoning_content` is being stripped or moved into the wrong field on the next turn. That would break agentic workflows, which makes the llama.cpp native UI a better reference implementation for now because it passes the chat-template kwargs through more directly. If you need the feature today, the likely fix is a pipe/filter or a targeted Open WebUI issue or PR rather than a hidden toggle.