Welcome to the "AI Daily" section! Here 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.
New AI products Click for more information:https://app.aibase.com/zh
1. JD Logistics launches "Super Brain Large Model 2.0" and "Yilang Embodied Intelligent Robotic Arm System"
JD Logistics announced the "Super Brain Large Model 2.0" and "Yilang Embodied Intelligent Robotic Arm System" at the JDDiscovery-2025 conference, marking a shift in the logistics industry from "supportive decision-making" to "embodied execution." These technologies enhance the efficiency of the logistics supply chain and achieve an intelligent closed-loop system.
【AiBase Summary:】
🚀 JD Logistics released "Super Brain Large Model 2.0" and "Yilang Robotic Arm," promoting the intelligent upgrade of the logistics industry.
📦 "Super Brain Large Model 2.0" can solve millions of variables within 2 hours, improving logistics operation efficiency.
🤖 The "Yilang" robotic arm has self-learning capabilities, enabling high-precision handling of complex packages.
2. DeepSeek V3.1 Terminus Version Released: Fixes Critical Bugs and Enhances Stability, V4 New Architecture on the Way
DeepSeek released the V3.1-Terminus version, significantly improving model stability and fixing critical issues. At the same time, this version may mark the end of the V3 series and pave the way for the upcoming V4 version or R2 major update.
【AiBase Summary:】
✅ Improved model stability, fixed issues with mixed Chinese and English output and abnormal characters.
🛠️ Optimized Code Agent and Search Agent modules, improving code generation and search accuracy.
🚀 Indicates that the V4 version or R2 major update is coming, possibly with a new architecture design.
3. Kimi Launches New Agent Mode "OK Computer", Begins Beta Testing
Kimi launched a new Agent mode called 'OK Computer' and began beta testing. This mode is based on the Kimi K2 model, supports multi-functional smart services, and users can complete complex tasks such as website development and data analysis simply by issuing requirements.
【AiBase Summary:】
🌟 Kimi launched a new Agent mode 'OK Computer', supporting multi-functional smart services.
💻 Users can complete complex tasks such as website development and data analysis by simply giving a request.
🚀 Kimi K2 model demonstrates excellent performance in autonomous programming and tool calling, with a total parameter count of 1T.
4. ChatGPT Launches Personalized News Feature, Your Personal News Assistant is Here!
ChatGPT introduced a personalized news feature, providing users with customized daily news based on their conversation history, covering multiple fields, and improving the efficiency of information retrieval.
【AiBase Summary:】
🤖 ChatGPT pushes personalized news based on user chat history.
📰 News covers areas like news, technology, and economy.
💡 Aimed at improving user's information retrieval efficiency and experience.
5. LLM Coding Hallucination Terminator! Exa Code Launches, Billions of Code Indexes Make AI Agents Error-Free
Exa Code is a network context tool optimized for Coding Agents, providing efficient and accurate code context by indexing a large number of documents and code repositories, helping large language models avoid generating incorrect code. Its core innovation lies in accurately extracting relevant tokens, reducing hallucination risks, and performing well in multiple scenarios.
【AiBase Summary:】
🔥 Exa Code is the first dedicated solution designed for Coding Agents, enhancing the reliability of AI agents.
💡 Accurate context extraction ensures maximum information density, avoiding long outputs.
🚀 Free, open-source, and easy to integrate, suitable for various programming needs, accelerating the development process.
Details link: https://exa.ai/blog/exa-code
6. Meta Launches AI Video Generation Platform Vibes, Easily Create and Share AI Short Videos
Meta's Vibes platform provides users with convenient AI video creation tools, supporting various creation methods and allowing cross-platform sharing, enhancing users' creative experience.
【AiBase Summary:】
🎬 Vibes is a platform that allows users to easily create and share AI-generated short videos.
🎨 Users can use existing materials, start from scratch, or remix dynamic content for creation.
📱 Videos can be directly published on Vibes or shared across platforms to Instagram and Facebook.
Details link: https://about.fb.com/news/2025/09/introducing-vibes-ai-videos/
7. Ant Group Launches New Privacy-Preserving AI Algorithm, Inference Speed Increased Over 100 Times
Ant Group achieved a major breakthrough in the field of privacy computing with its newly launched privacy-preserving AI training framework Gibbon, significantly improving the balance between data privacy and efficient computation.
【AiBase Summary:】
🧠 The Gibbon framework improves the training speed of traditional GBDT models through an innovative secure two-party training method.
🔒 Privacy-preserving decision tree inference technology based on homomorphic lookup tables greatly enhances computational efficiency under data privacy protection.
🌐 Ant Group has built a diversified privacy computing product matrix, providing high-security and high-performance solutions for multiple industries.
8. OpenAI's Latest Benchmark Test Shows GPT-5 Gradually Approaching Human Experts in Multiple Industries
OpenAI released a new benchmark test GDPval to evaluate AI models' performance in multiple industries compared to human professionals. GPT-5 and Claude Opus4.1 show close performance to industry experts in some tasks, but they still cannot fully replace human work. The test covered nine major industries and 44 professions, and plans to launch more comprehensive tests in the future.
【AiBase Summary:】
🌟 OpenAI's new benchmark test GDPval evaluates AI models' performance in multiple industries, gradually approaching the capabilities of human experts.
🤖 In 44 professions, GPT-5 performs better than or equal to industry experts in 40.6% of tasks, while Claude Opus4.1 achieves 49%.
📈 OpenAI plans to launch more comprehensive tests in the future to more accurately assess AI's capabilities and performance in real-world jobs.
Details link: https://openai.com/index/gdpval/
