XPENG Unveils World Model Blueprint and VLA 2.0 Mass Production at CVPR 2026
over 50%
4,360%
70%
What Happened
At CVPR 2026 in Denver, Dr. Xianming Liu, Head of General Intelligence Center at XPENG, delivered a keynote at the inaugural Workshop on Deployment of Foundation Models for Embodied AI, sharing the stage with leaders from Tesla, NVIDIA, and Waymo. He unveiled XPENG's complete technical blueprint for its Physical-World Foundation Model, highlighting breakthroughs in Deliberative Reasoning, Controllable Generation, and Long-Horizon Forecasting.
XPENG's VLA 2.0 ADAS, built on its self-developed foundational model, has entered formal mass production and achieved an industry milestone of over 50% assisted driving mileage share within its first month of OTA rollout. Designed from the ground up for Level 4 autonomous driving, VLA 2.0 enables a unified software architecture spanning both L2 and L4 capabilities.
4,360%surge
Over the 12 months ending March 2026, XPENG's cluster also delivered a 1,010% uplift in per-GPU training efficiency and GPU hardware utilization climbed from 40% to 90%.
“The successful launch of the initial VLA2.0 build has validated the capability gains brought by scaling up dataset volume and model parameters, further cementing our belief in the Scaling Law for physical-world AI.”
- VLA 2.0 ADAS
- Robotaxi (mass-produced, 3,000 TOPS computing power)
- Humanoid Robots (IRON, targeting mass production by end of 2026)
Why this matters
XPENG's advances in Physical AI and world models demonstrate how AI is transforming autonomous driving and robotics, with real-world scaling and efficiency gains validated by a 4,360% surge in training efficiency.
Terms in This Story
- Physical AI
- Artificial intelligence that interacts with and understands the physical world, enabling applications like autonomous driving and robotics.
- World Model
- An AI system that learns the underlying laws of the physical world to predict future states and enable reasoning and forecasting.
- VLA 2.0
- XPENG's second-generation Vision-Language-Action foundation model for advanced driver assistance, designed for Level 4 autonomy.
- Scaling Law
- The principle that increasing model size, data, and compute leads to predictable improvements in AI performance.
Related coverage
- XPENG launches China's first mass-produced Robotaxi in Guangzhou
- XPENG Unveils X-Cache World Model Accelerator, Boosting Inference Speed by 2.7x
- Schaeffler and Sonatus partner to integrate AI algorithms into control units for software-defined vehicles
- Hyundai Motor Group’s AI Roadmap: Minwoo Park on Commercializing AI and Autonomous Driving Technology
- Bosch Pushes Ahead with Key Technologies for Automation and Robotics at BCW 2026