XPeng to Scale AI Model to 20 Billion+ Parameters
XPeng is making a significant leap in its autonomous driving capabilities by scaling its on-vehicle AI model to over 20 billion parameters by the end of the year. Chairman and CEO He Xiaopeng announced the plan during the company's earnings call on March 20, confirming that its second-generation Visual Language-Action (VLA) model has already validated its scaling potential in real-world tests. This move from a model with several billion parameters to one exceeding 20 billion represents a major investment in computational power and AI sophistication, aiming to create a more human-like driving intelligence.
System Upgrade Targets 25x Increase in Autonomous Mileage
The massive parameter expansion is not just a technical benchmark but is directly tied to concrete performance goals. XPeng expects the new model to deliver a 25-fold improvement in the average mileage between driver interventions. More critically, the company projects a 50-fold increase in mileage before a safety-related takeover is required. These metrics are crucial indicators of system reliability and are designed to build consumer and regulatory confidence by drastically reducing the need for human oversight during autonomous operation.
Robotaxi Demonstrations to Launch in H2 2024
To translate its technological advancements into a viable business, XPeng plans to initiate passenger demonstration operations for its Robotaxi service in the second half of this year. This launch will serve as the first major public trial of the enhanced VLA model, providing crucial data on its performance in complex urban environments. The Robotaxi initiative is a critical step in XPeng's strategy to commercialize its autonomous driving stack, creating a new potential revenue stream and reinforcing its competitive position against rivals in the electric vehicle and self-driving sectors.
Edgen Stock·Mar 20 2026, 12:38
