I tested Qwen3 Coder Next against four other local AI coding models, and the gap was embarrassing
…Every other model either ignored the instruction entirely (Devstral 2), searched for the wrong thing (Qwen3.5 searched for generic "Python"), got the library IDs and then decided it didn't need…