中国基金报·2025-07-27 09:53Core Viewpoint - JD.com has launched the industry's first 100% open-source enterprise-level intelligent agent, JoyAgent, during the WAIC 2025, showcasing its advancements in AI applications across various industries [2][8][20]. Group 1: JoyAI Model Upgrade - The JoyAI model has been upgraded to include sizes ranging from 3 billion to 750 billion parameters, catering to diverse industrial needs across multiple modalities such as language, voice, image, and video [5][7]. - JoyAI's innovative technologies, including dynamic layered distillation and cross-domain data governance, have improved inference efficiency by an average of 30% and reduced training costs by 70% [5][7]. Group 2: JoyAgent Intelligent Agent - JoyAgent is the first fully open-source enterprise-level intelligent agent, featuring a product-level open-source structure that includes front-end, back-end, framework, engine, and core sub-agents [8][10]. - JoyAgent boasts three core features: 100% open-source capabilities, high availability with pre-configured sub-agents, and strong performance validated by over 20,000 internal use cases, achieving an accuracy rate exceeding 75% on the GAIA leaderboard [10][11]. Group 3: JoyInside Smart Interaction - JoyInside integrates role-based large models into smart hardware, enhancing product interaction experiences with high emotional intelligence and a vast knowledge base [13][15]. - The integration of JoyInside in educational tools has significantly increased user engagement, with average dialogue rounds rising by 148% and conversation duration improving by 50% [15]. Group 4: Digital Human Applications - JD.com has introduced high-commercial-use digital humans that outperform 80% of real hosts in emotional expression and scene adaptability, serving over 20,000 brands [18][19]. - The application of digital humans has expanded beyond live streaming to include brand representation and emotional interaction, significantly enhancing brand value and user experience [19][20]. Group 5: Industry Impact and Future Outlook - The deep application of large models is reshaping AI productivity, with JD.com demonstrating that effective AI is not just developed but also applied in real-world scenarios [20]. - The industry consensus is shifting towards application-level breakthroughs, indicating that the future of AI competition will focus on practical implementations rather than just foundational model development [20].