Li Auto Unveils MindVLA-o1 Model

[Technology Launch] At NVIDIA GTC 2026, Li Auto introduced its next-generation autonomous driving foundation model, MindVLA-o1.

Core Development: From Dual Systems to a Unified Multimodal Agent

MindVLA-o1 integrates 3D spatial understanding, multimodal reasoning, and behavior generation into a single native framework, replacing the previous end-to-end + VLM dual-system architecture. It also introduces closed-loop reinforcement learning and co-design of software and hardware.

Key Metrics: VLA Driver Model Already Deployed at Scale

As of the end of 2025, the previous-generation VLA driver large model achieved a monthly usage rate of 80%, processed over 12 million commands cumulatively, and supported a total assisted-driving distance of 250 million kilometers during the Spring Festival holiday period.

Strategic Foundation: Advancing Toward General Intelligence in the Physical World

This model is no longer limited to autonomous driving; instead, it serves as the foundation for a general-purpose agent operating in the physical world, marking Li Auto’s evolution from functional intelligent driving toward embodied intelligence.