Qwen3 is spotty with tools at the moment. gpt-oss-120b is better with it