Qwen3.6 think tags stick in prompts

This looks less like a broken model and more like a runtime mismatch with Qwen3.6’s newer thinking template. The official docs show Qwen3.6 thinks by default and expose separate `enable_thinking` and `preserve_thinking` controls, so output parsing alone will not rewrite how history is fed back in.

• –Qwen’s docs say Qwen3.6 operates in thinking mode by default and that `preserve_thinking` can feed prior reasoning back into the next turn.
• –`--reasoning-format deepseek` can strip what downstream consumers see, but it does not necessarily change the tokenizer/chat-template state that gets reinserted upstream.
• –If Unsloth behaves and APEX does not, the difference is probably in template implementation, not model quality.
• –For agents and tools like Frigate, the real fix is likely a nonthinking or custom chat template, or explicit prompt cleanup after template application.
• –This is the kind of change that breaks integrations silently: the assistant output looks fine, but the next-turn context still carries hidden reasoning scaffolding.