Welcome to the "AI Daily" section! This is your guide to exploring the world of artificial intelligence every day. Every day, we present you with the latest content in the AI field, focusing on developers, helping you understand technical trends and innovative AI product applications.

Fresh AI products click to learn more:https://app.aibase.com/zh

1. Beijing releases the "Artificial Intelligence Industry White Paper (2025)", expecting core output value to exceed 450 billion yuan

The article details the holding of the 2025 China Artificial Intelligence Conference in Beijing, as well as the release of the "Beijing Artificial Intelligence Industry White Paper (2025)" by the Beijing Municipal Science and Technology Commission. The white paper analyzes the current status of global and Chinese artificial intelligence development and points out that the scale of the artificial intelligence industry in Beijing has grown rapidly, expected to exceed 450 billion yuan by the end of 2025.

image.png

【AiBase Summary:】

🌍 Analysis of global artificial intelligence development trends, expecting core output value to exceed 450 billion yuan by 2025.

🚀 Beijing has become a gathering place for AI enterprises nationwide, with over 2,500 companies.

💡 Through policy guidance and scientific research innovation, Beijing continues to promote high-quality development of the AI industry.

2. ByteDance launches the "Shaking-Level" AI model Vidi2: 12 Billion Parameters, Making Video Editing Fully Automated

ByteDance has launched the multimodal large language model Vidi2, with 12 billion parameters, focusing on video understanding. The model can process long videos and generate short videos or movie clips based on simple prompts, with precise spatiotemporal positioning functions, improving the efficiency and accuracy of video editing.

image.png

【AiBase Summary:】

🎥 Vidi2 has precise spatiotemporal positioning (STG) functionality, capable of identifying timestamps and object bounding boxes in videos.

⚙️ The model uses Gemma-3 as the backbone network and combines adaptive token compression technology to ensure efficiency and detail retention when processing long videos.

📱 Tools developed based on Vidi2 have been applied to TikTok, such as Smart Split, achieving automatic editing, reconstruction, and adding subtitles.

Details link: https://www.alphaxiv.org/abs/2511.19529

3. Tibet's First Trillion-Parameter Tibetan Language Large Model "Sunshine Qingyan" Released, Marking a New Stage for AI on the Plateau

Tibet's first trillion-parameter Tibetan language large model "Sunshine Qingyan" was released, marking a new stage for AI on the plateau, providing important support for the development of Tibetan AI.

image.png

【AiBase Summary:】

🧠 Tsinghua University and a startup team jointly released "Sunshine Qingyan" V1.0, with parameter count exceeding a trillion, covering multi-domain corpus.

🌍 Tibetan AI infrastructure has achieved breakthroughs, providing a plateau sample for low-resource language intelligence.

🏥 Government services and remote medical care scenarios will achieve mother-tongue interaction, improving the efficiency of Tibetan language inheritance.

4. Kuaishou Kling Omni to Be Launched This Week: Achieving "Director-Level" Precision Control, Generating 2-Minute Videos with Native Audio

Kling AI, a subsidiary of Kuaishou Technology, officially launched 'Kling Omni Launch Week', introducing the multimodal video foundation model Kling O1 (Omni One), which supports multiple input references including text, images, and videos, offering director-level control capabilities. The model can generate continuous videos up to 2 minutes long, achieving frame-level precise audio synchronization, marking a shift in AI video creation from random generation to programmatic director mode.

image.png

【AiBase Summary:】

🎥 Kling O1 supports multiple input references including text, images, and videos, achieving director-level precision control.

⏱️ Can generate continuous videos up to 2 minutes long, enhancing motion stability and style consistency.

🔊 Integrates native audio synchronization technology, ensuring sound and image reach frame-level precise matching.

5. Meta AI Launches Matrix Framework, Revolutionizing Multi-Agent Synthetic Data Generation

Meta AI introduced the Matrix framework, solving the issues of freshness and diversity of synthetic data through a decentralized design, while also improving token throughput.

image.png

【AiBase Summary:】

🧠 The Matrix framework adopts a decentralized design, avoiding bottlenecks of traditional central schedulers.

🚀 In multiple case studies, Matrix showed a 2 to 15 times improvement in token throughput.

🔧 The framework fully utilizes the distributed features of Ray clusters to achieve efficient synthetic data generation and processing.

Details link: https://arxiv.org/pdf/2511.21686

6. Domestic Home Robot F1 Is Here: Push You to Work With Your Bed, Price Less Than 20,000 Yuan Next Year

The article introduces the innovative features and market plans of the domestic home robot F1. F1 has 22 joints, a wheeled chassis, and an 8-hour battery life, capable of performing various household tasks such as cleaning and child-rearing, and executing long sequences of起床 (rising) tasks. It uses the RVLA model architecture, with a task success rate exceeding 94%. In addition, F1 solves the problem of space limitations in small apartments, and will launch an educational module and plan to go on sale next Q1.

image.png

【AiBase Summary:】

🤖 F1 has 22 joints and a wheeled chassis, capable of performing various household tasks such as cleaning and child-rearing.

💡 F1 uses the RVLA model architecture, with a task success rate over 94%, and can automatically retry obstacles encountered.

🚀 F1 plans to launch next Q1, priced under 20,000 yuan, with the first 1,000 units available for purchase via WeChat Mini Program.

7. Douba Upgrades Voice Function! Supports Four Dialects, Helping the Elderly Communicate Without Barriers

Douba App now adds support for four local dialects in voice conversation functions, enhancing user interaction with AI, especially convenient for elderly people who are not familiar with Mandarin. In addition, Douba also demonstrated its continuous innovation in the technology field, such as using ASR technology to improve recognition accuracy, and hosting scholarship events to support tech talent development.

image.png

【AiBase Summary:】

🍊 Douba App now supports voice conversation functions in four dialects: Cantonese, Sichuan dialect, Northeastern dialect, and Shaanxi dialect.

🧠 The Douba voice model has dialect migration technology, capable of switching dialects flexibly according to user intent.

💰 ByteDance held a scholarship award ceremony, increasing the scholarship amount to 200,000 yuan to support the development of tech talents.

8. Douba Mobile Assistant Releases Technical Preview Version

The Douba team officially launched the technical preview version of the Douba mobile assistant, aiming to provide users with a more efficient interaction experience and more rich functions. The assistant relies on the powerful capabilities of the Douba large model and combines system authorization from mobile phone manufacturers, striving to create a brand-new user experience.

image.png

【AiBase Summary:】

🌟 The technical preview version of the Douba mobile assistant was officially released today, dedicated to improving user interaction experience.

📱 The nubia M153 engineering prototype equipped with the Douba mobile assistant has been sold in small quantities, targeting developers and interested users.

🚀 The Douba team will continue to optimize the assistant's functions and collaborate with multiple mobile phone manufacturers to promote the maturity of mobile assistants.

Details link: https://o.doubao.com/