At the Beijing Auto Show, Ruan Chong, a former core researcher of DeepSeek's multimodal technology, officially made his debut as the Chief Scientist of Yuanrong Qixing. This move marks a significant shift in Yuanrong Qixing's strategy for autonomous driving technology.

Yuanrong Qixing's CEO, Zhou Guang, stated on-site that multimodal large models have achieved breakthrough progress by early 2026. He pointed out that compared to the previous generation of technology, the autonomous driving approach based on large models has shown a dominant advantage from the start. The "small model" architecture widely used in the industry in the past suffered from an inescapable "see-saw effect" — optimizing performance for one scenario often harmed another, making it difficult to achieve full-scenario safety coverage. Based on this, Yuanrong Qixing has decided to fully bet on the large model approach.

Chief Scientist Ruan Chong further detailed the company's technical transformation. Currently, Yuanrong Qixing is working on transitioning from multiple scattered small model architectures to a unified base large model architecture. Under this new architecture, the system is divided into three sub-models with specific functions: driving, analysis, and commentary, thereby achieving more accurate decision-making and processing.

The improvement in technical efficiency is also remarkable. Thanks to the optimized architecture, the current single model iteration cycle has been significantly shortened, from over 100 hours to just over 10 hours. This order-of-magnitude increase in development speed will provide strong support for the rapid evolution and safe implementation of autonomous driving technology.