The WAVE SUMMIT Deep Learning Developers Conference, hosted by the National Engineering Research Center for Deep Learning Technology and Applications and jointly organized by Baidu PaddlePaddle and ERNIE Model, kicked off in Beijing. At the event, Baidu officially launched the ERNIE Model X1.1, alongside the release of PaddlePaddle Core Framework v3.2 and a series of technological, product, and ecosystem achievements, demonstrating China’s innovation capabilities in the field of artificial intelligence.
During the release, Baidu’s Chief Technology Officer and Director of the National Engineering Research Center for Deep Learning Technology and Applications, Wang Haifeng, stated that the ERNIE Model X1.1 is trained based on ERNIE 4.5 and adopts an iterative hybrid reinforcement learning framework. Through self-distillation data iteration production and training, the model’s performance has been significantly improved. Compared to the previous X1 version, X1.1 achieves a 34.8% improvement in factual accuracy, a 12.5% increase in instruction-following capabilities, and a 9.6% enhancement in agent interaction efficiency. The model is now fully available to individual users and enterprise developers through the ERNIE Bot official website, the Wenxiaoyan APP, and Baidu Intelligent Cloud’s Qianfan Platform.
In authoritative benchmark evaluations, the ERNIE Model X1.1 outperformed DeepSeek R1-0528 overall, demonstrating leading advantages in certain tasks. Compared to top international models such as GPT-5 and Gemini 2.5 Pro, it achieved comparable results. This achievement is attributed to the joint optimization of PaddlePaddle and ERNIE. The PaddlePaddle Core Framework v3.2 has undergone comprehensive upgrades in large model training, hardware adaptation, and ecosystem support. The simultaneously updated ERNIEKit development suite and FastDeploy deployment suite further lower the technical barrier.
At the conference, Baidu’s ERNIE Model, in collaboration with the China Aging Development Foundation, launched the “AI for the Elderly Public Welfare Initiative.” Through intelligent agents, AI applications, and public welfare volunteer activities, this initiative aims to help the elderly bridge the digital divide. Currently, the “AI for the Elderly Intelligent Agent” is already available, and users can access its services via Baidu Search or Xiaodu Smart Health Screen to experience the convenience brought by AI technology.
Baidu Group Vice President and Deputy Director of the National Engineering Research Center for Deep Learning Technology and Applications, Wu Tian, demonstrated practical application scenarios of the ERNIE Model X1.1. In complex writing tasks, the model can integrate internalized knowledge with internet search tools to generate well-structured, logically rigorous, and eloquently written content. When addressing multi-dimensional questions from shared bike platform users, the model can autonomously plan processes and invoke tools based on user levels, question types, and emotional states to efficiently complete services.
Leveraging the technological breakthroughs of the ERNIE Model, Baidu introduced script-driven multi-modal digital human technology. This technology integrates language, voice, and image generation, achieving digital human anchors with “high emotional intelligence, strong interactivity, long endurance, and greater professionalism” through five innovations: script generation, dynamic interaction, voice synthesis, and long video generation. In practical applications with Baidu Huiboxing, digital human livestreams have outperformed human hosts online. Luo Yonghao’s digital human livestream debut set a new industry record in GMV, with sales volumes for certain categories surpassing those of human hosts.
It’s reported by the latest news that Wu Tian also provided updates on the latest developments in the ERNIE data ecosystem. Within the ERNIE Mentor team, 48% of experts have over 10 years of industry experience, while 41% are professionals from technical and educational backgrounds. The launched “ERNIE Mentor · Starlight Plan” will leverage the resources of the PaddlePaddle-ERNIE ecosystem to collaborate with industry elites and domain experts in exploring innovative applications of AI technology. To date, the PaddlePaddle-ERNIE ecosystem has attracted 23.33 million developers and served 760,000 enterprises.
