Sora's Big Three Reveal Behind-the-Scenes Information for the First Time, CTO: Open Public Testing Expected This Year


The Tencent Hua yuan team has open-sourced the video generation acceleration solution DisCa. By using learnable feature caching technology, it further compresses the inference cost on the distilled few-step model and improves the generation speed. The code and model weights of this solution are publicly available and have been accepted by CVPR 2026.
According to Stanford University's 2026 AI Index Report, global AI technology has entered a period of full popularization, with the top technological gap between the US and China reaching a historical low. The United States maintains an advantage in top models, high-impact patents, and the number of data centers, but China has caught up in multiple core indicators, forming a competitive landscape where leadership alternates between the two countries.
OpenAI CEO Sam Altman stated that shutting down Sora is not due to technical issues but a severe shortage of computing power, as the company must allocate limited chip resources to higher-priority projects, leading to internal competition for resources.....
Kuaishou's Kling AI leads globally in mobile video models, gaining prominence after OpenAI's Sora shutdown, with 7.8 million MAUs far surpassing Sora's 4.7 million.....
Alibaba releases the Wan2.7-Image model, achieving breakthroughs in portrait customization, color control, and long-text rendering, aiming to solve the problem of 'monotonous' AI-generated images. The model has been launched on the Alibaba Cloud BaiLian platform API and is open for experience. Core upgrades include the 'virtual face sculpting' and 'precise color palette' features, significantly enhancing creative freedom.