Alibaba's Qwen-Drive-1.0-4B: open 4B driving VLM beats 3x larger Gemma on average score
eyishazyer · x · 2026-09-07
Alibaba's Qwen team, with Huazhong University of Science and Technology, released Qwen-Drive-1.0-4B, which they call the first vision-language foundation model built for driving. Key numbers:
- 69.43 average driving VQA score, beating the 3x larger Gemma4-12B (58.15)
- 90.7 on NAVSIM planning, the highest among tested models
- 43.95 mAP on nuScenes perception, competitive with dedicated 3D detection systems
- Built on the Qwen3.5-4B backbone with a perception head and planning module for trajectory generation
Notably, two planner versions were shipped: imitation-trained and RL-tuned. The RL version halved the off-road rate while becoming more cautious — safer instincts baked into the weights. Everything is released under Apache 2.0 with code, weights, and demo data.
Related event: Alibaba Open-Sources Qwen-Drive-1.0-4B for Autonomous Driving(3 posts)→
More from Embodied
- New tendon-based robotic hand designs trained with PPO to learn one-armed catch — _krishna_murthy · 2026-10-23
- Rivian acquires world model startup Third Dimension to accelerate its self-driving push — emmanuelvivier · 2026-10-23
- Nvidia's Berlin GTC spotlights humanoid robots and Vera Rubin AI supercomputing — emmanuelvivier · 2026-10-23
- XPeng commissions 80%-automated line for IRON humanoid, mass production by end-2026 — ingliguori · 2026-10-23
- HKU's TILDE Distills Embodied Foundation Models into Robot Policies — hkuhk · 2026-10-23
- 10 of 12 Beta Robots Worked Without Fine-Tuning — Do We Still Need It? — micoolcho · 2026-10-23