No, exactly the opposite actually. Qwen3.6 is too imprecise for long running agentic tasks. It doesn't have the same ability to check itself as Gemma does in my testing. I keep Qwen MoE in vram by default because there are tons of tasks i trust it to oneshot and it's 90tok/sec is unparalleled, anything where I don't want to have to intervene too much it can't be trusted.