信息服务行业深度报告：AI商业化的旗手：不止豆包的字节跳动

Industry Investment Rating - The report does not explicitly provide an overall industry investment rating, but it highlights the strong performance and growth potential of ByteDance's AI models, particularly the Doubao model, which has seen significant adoption and usage across various industries [3][21][29] Core Viewpoints - ByteDance's Doubao model has achieved widespread recognition for its capabilities and inference effects, with daily token usage exceeding 4 trillion, a 33-fold increase in 7 months [29][30] - The model's low pricing strategy, with prices as low as 0.0008 yuan per thousand tokens, has made it accessible to a wide range of enterprises, driving rapid adoption and usage growth [53][145] - ByteDance's AI models are being applied across multiple industries, including finance, automotive, retail, and education, with significant growth in token usage, especially in emerging sectors like automotive and smart devices [29][31][34] Model Capabilities and Development - Doubao model has been continuously upgraded, with the latest version Doubao-pro achieving performance comparable to GPT-4o at a fraction of the cost [200][205] - The model supports multi-modal capabilities, including text, image, video, and audio, with applications in real-time voice communication, video generation, and visual understanding [70][72][74] - ByteDance has developed a comprehensive ecosystem of AI tools and platforms, such as Volcano Engine and Coze, to facilitate the deployment and integration of AI models across various industries [15][131][148] Industry Applications - Doubao model has been widely adopted in industries such as finance, automotive, retail, and education, with companies like China Merchants Bank and Mengniu leveraging the model for various AI-driven applications [28][34] - The model has seen particularly strong growth in the automotive sector, with token usage increasing by over 50 times since its launch [29][34] - ByteDance has formed industry-specific alliances, such as the Automotive Big Model Ecosystem Alliance and the Retail Big Model Ecosystem Alliance, to promote the adoption of AI models in these sectors [31][34] Technological Innovations - ByteDance has introduced several technological advancements, including the HybridFlow framework for reinforcement learning and the FAN (Fourier Analysis Networks) architecture for periodic modeling [177][142] - The company has also developed specialized models for tasks such as video generation, visual understanding, and music composition, which have been integrated into the Doubao ecosystem [72][74][87] - ByteDance's AI models are supported by a robust infrastructure, including high-performance computing resources and advanced networking capabilities, enabling efficient model training and deployment [163][164]