AI Daily: MiniMax Launches M3 Large Model; NVIDIA's Physical Large Model Cosmos3 Released; Xiaohongshu Has Removed Over 1.2 Million AI-Managed Accounts

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 to help you understand technical trends and learn about innovative AI product applications.

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

1. MiniMax releases M3 large model: First to adopt MSA architecture and support 1M context, fully open source to match overseas top-tier models

The release of MiniMax's M3 large model marks a major breakthrough in the fields of multimodal and long-context technology in China. Its innovative MSA architecture and open-source strategy bring new opportunities to the developer ecosystem.

【AiBase Highlights:】

🧠 The M3 model uses a self-developed sparse attention architecture (MSA), significantly improving computing speed and efficiency.

🚀 The M3 achieves only one-tenth the per-token computation of the previous generation model under 1M context, with significant acceleration in pre-filling and decoding stages.

🌐 The M3 surpasses GPT-5.5 and Gemini3.1Pro in authoritative evaluations, demonstrating strong long-thread autonomous planning capabilities.

2. Cosmos3, the world's first fully open-source, multi-modal physical AI large model, is launched, with NVIDIA joining industry giants to establish the Cosmic Alliance

The article introduces NVIDIA's launch of the world's first fully open-source, multi-modal physical AI large model, Cosmos3, which has a disruptive breakthrough in the field of physical AI, and has formed the Cosmic Alliance with industry leaders to promote technological development.

【AiBase Highlights:】

🧠 Cosmos3 is the world's first fully open-source, multi-modal physical AI large model, capable of visual reasoning, world generation, and action prediction.

🧪 The model is trained on a vast physical AI dataset and can naturally understand and generate cross-modal content with industry-leading simulation accuracy.

🚀 NVIDIA has joined forces with multiple top teams to establish the "Cosmic Alliance," driving the transformation and development of physical AI.

3. Qwen APP's "Gift Selection" inquiry volume increased by nearly 260% in the past week

The article analyzes the surge in AI shopping demand during Children's Day, pointing out the important role of Qwen APP in gift selection, and emphasizing its ability to enhance user experience through intelligent recommendations and full-process services.

【AiBase Highlights:】

🍎 During Children's Day, AI shopping demand significantly increased, with a sharp rise in user inquiries about gift selection.

💡 Qwen APP provides precise product recommendations by understanding user needs, improving shopping efficiency.

🛒 Qwen is fully integrated with Taobao, enabling a complete shopping process from recommendation to after-sales service.

4. Major crackdown! Xiaohongshu has dealt with over 1.2 million AI-managed accounts and 180,000 AI-fabricated notes

Xiaohongshu shared its latest practices and insights on AIGC creation, community governance, and maintaining the ecosystem for emerging creators at the 2026 Shanghai Network Audiovisual Creators Conference. The platform has cumulatively handled 180,000 AI-fabricated notes and systematically managed over 1.2 million AI-managed matrix accounts to maintain the authentic sharing and trustworthy interaction ecology of the community.

【AiBase Highlights:】

🤖 Handling of AI-fabricated notes: Xiaohongshu has dealt with 180,000 AI-fabricated notes.

👥 Cleanup of AI-managed accounts: The platform has systematically managed over 1.2 million AI-managed accounts.

🛡️ Community ecological maintenance: By upgrading model identification and tiered governance, the platform enhances its ability to detect and distinguish AI content.

5. Over 44,000 AI short videos added on Douyin in April, 13 times that of real actor dramas, with only 0.6% of them achieving over 100 million views

The article points out that AI short videos have rapidly risen on platforms like Douyin, but the market has a high elimination rate, with rare high-quality content, and future competition will focus more on the integration of creativity and technology.

【AiBase Highlights:】

🎥 AI short video production has surged, with over 44,200 new works added on Douyin in April, 13 times that of real actor dramas.

📉 High market elimination rate, with only 0.6% of AI videos achieving over 100 million views, indicating a scarcity of hit content.

🚀 Future competition will shift from efficiency-driven to refined quality battles involving deep integration of creativity and technology.

6. From Musician to "Vibe Coder", Hu Yanbing Launches Fan Community App "Yanhuo"

Hu Yanbing showcased the potential of Vibe Coding through the "Yanhuo" app, a new development paradigm that allows creative workers to overcome coding barriers and manage the entire lifecycle of an application from concept to implementation.

【AiBase Highlights:】

🎧 Hu Yanbing launched the fan community app "Yanhuo," marking the breaking of technical barriers by the Vibe Coding development paradigm.

💻 The app includes features such as check-ins, points, and letters, and will expand to include Q&A and dynamic updates in the future.

🧠 Vibe Coding converts natural language into productivity, allowing non-professional developers to realize application development.

7. LobsterAI launches image and video large model matrix, integrating four mainstream image and video generation models at once

LobsterAI has made significant progress in image generation and video generation, providing users with more diverse visual content creation support by integrating four mainstream multi-modal large models, and reducing the usage threshold, promoting further development of digital content production.

【AiBase Highlights:】

🧠 LobsterAI officially launched image and video generation capabilities, offering users more diverse visual content creation support.

🧩 Integrated four mainstream multi-modal large models at once, including Seedream, Seedance, HappyHorse, and MiniMax-Hailuo.

🚀 Through a multi-model matrix collaboration mode, it lowers the threshold for developers and creators to access multi-modal AI capabilities, promoting the development of digital content production.

8. OpenAI Announces Restart of Robot Business, Omotenashi Publicly Recruits Engineers

OpenAI CEO Sam Altman announced entering the robot market, restarting the embodied intelligence layout in the physical world, aiming to expand the capabilities of large models from the digital world to the real world.

【AiBase Highlights:】

🤖 OpenAI announced the restart of its robot business, recruiting full-stack hardware, operations, systems, and machine learning engineers.

🏗️ Short-term goal is to develop specific function robots that assist technical workers in building future infrastructure.

🌐 Long-term vision is to create versatile personal robots that meet diverse customization needs.