Qwen3 4B

This qwen3 hybrid reasoning model enables seamless switching between thinking and non-thinking modes during conversations, achieving SOTA reasoning performance at its scale. It shows significant improvements in human preference alignment, creative writing, role-playing, multi-turn dialogue, and instruction following, delivering a greatly enhanced user experience.