Sou Hu Cai Jing·2025-04-27 01:21Core Insights - Baidu held its annual AI developer conference, Create 2025, in Wuhan, showcasing its advancements in AI models and applications [1] - The focus of the conference was on the importance of applications over models, emphasizing that true value comes from practical applications [4] Model Developments - Baidu introduced two new models: Wenxin 4.5 Turbo and Wenxin X1 Turbo, highlighting their capabilities in image and video understanding [1][2] - Wenxin 4.5 Turbo demonstrated its ability to identify historical sports events from low-quality images, showcasing its advanced analytical skills [1] - Wenxin X1 Turbo exhibited deep thinking and multi-tool capabilities by generating a unique representation of a local cultural symbol, the "garlic bird" [2] Cost Efficiency - The cost structure for Wenxin 4.5 Turbo is significantly reduced, with input costs at 0.8 yuan per million tokens and output costs at 3.2 yuan, marking an 80% decrease compared to its predecessor [3] - Wenxin X1 Turbo has input costs of 1 yuan and output costs of 4 yuan per million tokens, reflecting a 50% reduction in costs while improving performance [3] - Lower costs are expected to encourage developers and entrepreneurs to create AI applications, facilitating widespread adoption across various industries [3] Industry Applications - Baidu is collaborating with the China Cultural Relics Exchange Center to enhance the model's expertise in cultural heritage, launching the first cultural heritage intelligent agent [3] - In the field of intangible cultural heritage, Baidu partnered with Shanghai University of Sport and the Chinese Martial Arts Museum to digitize martial arts techniques using AI [3] - Additional initiatives include the launch of persuasive digital humans, the Cangzhou OS system, and a multi-agent collaboration app, along with plans to support developers in embracing the MCP [3] Market Trends - The rapid increase in the number of AI model releases was noted, with 49 models launched in Q4 of the previous year and 55 in Q1 of 2025, indicating a competitive landscape [4] - The emphasis remains on creating valuable applications, as models and chips alone do not generate value without practical use cases [4]