Qwen turns a compact vision model toward driving decisions
Qwen is taking its 4B-class multimodal model out of generic image understanding and into driving-oriented planning. The official checkpoint exposes a dedicated `QwenDriveForPlanning` interface and runs through Transformers, vLLM, or SGLang, making it unusually easy to test or integrate. The catch: this is a model release, not evidence of a production-ready autonomous-driving stack.