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 technology trends and innovative AI product applications.
New AI products click to learn more: https://app.aibase.com/zh
1. Kimi k2 performance praised to exceed GPT-5, Moonshot AI secures another round of billions in funding
The domestic large model startup company Moonshot AI is about to complete a new round of funding worth several hundred million dollars. This financing comes shortly after the previous round of about 300 million dollars (the last public information was in August 2024), showing increased confidence from the capital market in the company.
AiBase Summary:
🚀 Kimi k2 model performance exceeds GPT-5 and Claude4.5, attracting global attention.
💰 Moonshot AI completes billions in funding, with high confidence from the capital market.
🌐 The release of Kimi k2 has once again made Moonshot AI a focal point on the global AI stage.
2. Anthropic introduces 'memory' feature for Claude Pro/Max users, enhancing personalization and security
Anthropic has launched a new 'memory' feature for its flagship AI model, Claude Pro and Max versions, aiming to enhance the consistency and personalization of conversations. This feature emphasizes privacy and isolation, allowing users to control their memory content and offering a 'private chat' mode to ensure data is not saved.
AiBase Summary:
🧠 The memory function can remember project content, user preferences, and workflow, improving conversation coherence.
🔒 Project-specific memory and user control design ensure privacy and data isolation.
🕵️♂️ New 'private chat' mode ensures that conversation data is not saved or displayed in records.
3. University of Science and Technology of China and ByteDance release MoGA long video generation model: generate multi-shot short films in minutes
University of Science and Technology of China and ByteDance jointly released an end-to-end long video generation model using the MoGA algorithm to achieve high-quality video generation within minutes, marking a key breakthrough for domestic technology in the global AI competition.
AiBase Summary:
🎥 Generate multi-shot videos within minutes, breaking traditional time limits.
🧠 Introduce MoGA algorithm to optimize context processing and computational cost.
🚀 Support multiple fields such as film and advertising, with potential for industrial application.
More details: https://jiawn-creator.github.io/mixture-of-groups-attention/
4. Tencent releases new ima2.0: task mode added, new 'AI highlights' features
Tencent officially released the upgraded ima2.0 version at the ima Open Day event, introducing the 'task mode', which upgrades the knowledge base from a simple search and question answering tool into an intelligent partner that can understand complex tasks, autonomously break them down, call tools, and complete the entire process. In addition, the new 'AI highlights' feature supports parallel tasks and collaborative sharing, improving the efficiency of knowledge management.
AiBase Summary:
🔥 ima2.0 introduces the 'task mode', enabling automatic decomposition and execution of complex tasks.
💡 New 'AI highlights' feature automatically generates structured summaries, quickly capturing key points.
🤝 Supports parallel tasks and collaborative sharing, improving the efficiency of knowledge base usage.
5. Alibaba Quark AI Glasses launch pre-orders: support Alipay "Look" payment
Alibaba Quark AI Glasses are launched, supporting Alipay payments, featuring excellent design and powerful photography capabilities to meet diverse user needs.
AiBase Summary:
📱 Supports Alipay "Look" payment, becoming the first smart glasses in China to integrate biometric payment.
👓 Features 7.5mm ultra-thin temples and ultra-thin frames, weighing only 42 grams, comfortable to wear.
📸 Equipped with SuperRaw night vision mode and dual anti-shake system, enhancing photography capabilities.
6. Microsoft Copilot adds group chat, memory, and Edge AI mode: up to 32 people can collaborate in real-time
Microsoft made a major fall update to Copilot, adding group chat, long-term memory, and Edge AI mode, enhancing collaboration, personalization, and cross-service integration.
AiBase Summary:
🌟 Introduces "group" chat function, supporting real-time collaboration and task assignment for up to 32 people.
🧠 Adds "long-term memory" function, saving user key information and automatically calling it in subsequent conversations.
🌐 Edge browser launches Copilot mode, providing content analysis and smart suggestions.
7. Opera Neon browser launches deep research Agent "ODRA"
Opera Neon browser launched a new AI feature - Opera Deep Research Agent (ODRA), marking a significant progress in Opera's AI ecosystem development. ODRA improves the efficiency and accuracy of complex query through parallel processing, demonstrating its competitiveness against industry giants.
AiBase Summary:
🌍 ODRA is the core part of Opera's self-developed AI engine, with significantly improved performance after more than two years of R&D and optimization.
⚙️ ODRA uses parallel operations, splitting complex tasks into multiple sub-problems to improve research efficiency.
🔍 ODRA is the fourth agent added to Opera Neon, and users can start the research function through the all-in-one search bar.
8. Baidu PaddleOCR-VL model tops the global OCR ranking, leading Huggingface trend list for five consecutive days
The PaddleOCR-VL model released by Baidu's Paddle team has caused a sensation in the OCR field, becoming the most popular open-source OCR model due to its strong performance and multilingual support.
AiBase Summary:
✅ PaddleOCR-VL scored 92.56 in the OmniDocBench V1.5 evaluation, surpassing mainstream models and topping the global OCR rankings.
🌐 Supports 109 languages, capable of parsing text, tables, formulas, and charts, and possesses the ability to reconstruct document semantic structure.
🔍 Institutions such as Baidu, DeepSeek, and Shanghai AI Lab have open-sourced OCR models, with the core goal of providing data foundation for large model training.
