Qwen open-sourced Qwen-Drive-1.0-4B, a unified vision-language model for autonomous driving
Qwen released Qwen-Drive-1.0-4B on Hugging Face under Apache-2.0. The model unifies 3D perception, visual question answering, and motion planning for autonomous driving, retaining the Qwen3.5-4B VLM architecture.
China context
- Original name
- 通义千问
- Outside China
- Open weights · huggingface.co
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can download and fine-tune the model for autonomous driving tasks under Apache-2.0, but must verify performance on their own data and comply with local regulations for autonomous vehicle deployment.
- For investors
- The release of an open-weights driving model by Alibaba may pressure other Chinese AI labs to open their driving models, affecting the competitive landscape and potential partnerships in the autonomous driving sector.
Qwen-Drive-1.0-4B is a vision-language foundation model for autonomous driving that integrates 3D perception, visual question answering, and motion planning within a unified framework. It retains the architecture of the pretrained Qwen3.5 vision-language model and adds a BEV perception head and a Planning Expert. The model is released on Hugging Face under Apache-2.0 license.
The model uses a natively multimodal Qwen3.5-4B as the shared VLM, with an external BEV perception head for 3D object detection, semantic occupancy prediction, and BEV map segmentation, and a Planning Expert that generates future ego trajectories through flow matching. Two Planning Experts are released: planner-sft and planner-rl, with planner-rl optimized on NAVSIM PDMS, WOD-E2E RFS, and a displacement term.
Autonomous driving developers outside China gain access to an open-weights model that combines perception and planning, potentially reducing the cost of building driving stacks compared to proprietary alternatives. Competitors must now match or exceed its open-loop and closed-loop planning benchmarks to remain credible.
For developers, the Apache-2.0 license allows commercial use and modification, lowering barriers to entry for autonomous driving research and product development. For investors, the release signals Alibaba's commitment to open-weight models in the competitive autonomous driving AI space.
The next observable signal is whether Qwen-Drive-1.0-4B is adopted in downstream driving simulators or deployed in real vehicles, and whether independent benchmarks confirm the reported WOD-E2E RFS of 8.45 for the RL planner.