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 learn about innovative AI product applications.
Fresh AI products Click to learn more: https://app.aibase.com/zh
1. Alibaba's New Large Model Qwen 3.8 Is Coming: Preview Version Launched on Alibaba Cloud and Qoder, Official Version to Be Open-Sourced Soon
Alibaba's new large model Qwen 3.8 will be released and open-sourced simultaneously. The preview version has been launched on multiple platforms, allowing users to experience its capabilities in advance and test the model's performance through real user scenarios.
【AiBase Highlights:】
🚀 Alibaba is about to release the new large model Qwen 3.8 and will open-source it simultaneously, further advancing the open-source process of the Tongyi Qianwen family.
🔧 The Qwen3.8-Max preview version has been launched on multiple platforms, allowing users to experience the new model's capabilities in advance.
📈 Alibaba chooses to release the preview version first and follow up with the official version for open-source, which not only conveys product signals but also accepts testing under real user scenarios.
2. ByteDance Releases Seed Audio 1.0: Audio Generation Moves from "Speaking" to "Creating"
ByteDance's Seed Audio 1.0 marks a new stage in AI audio generation technology, capable of generating complete audio works end-to-end. The model features core capabilities such as precise temporal and spatial scheduling, stable voice expression, and multilingual support, significantly lowering the barrier to high-quality audio content creation.
【AiBase Highlights:】
🌱 Precise Temporal and Spatial Scheduling: Supports controlling the timing of dialogue and sound effects with 100-millisecond precision.
🎭 Stable Voice Expression: Supports zero-shot generation and long audio extension, naturally presenting different emotions.
🌍 Authentic Multilingual Support: Covers over 20 languages, ensuring voices match local expressions and rhythms.
More Details: https://seed.bytedance.com/seedaudio1_0
3. Mingshi Intelligence Opens MiniCPM-Robot: A 1.5B VLA Model Fully Released, Personal Developers Can Run It on Real Robots
Mingshi Intelligence has open-sourced the first embodied AI model series, MiniCPM-Robot, including a 1.5B parameter visual-language-action model that allows personal developers to run operations and target tracking on real robots. The model series includes MiniCPM-RobotManip, MiniCPM-RobotTrack, and the high-performance inference framework PhyAI, providing important support for the popularization of embodied intelligence.
【AiBase Highlights:】
🧠 MiniCPM-RobotManip: A 1.5B parameter general-purpose visual-language-action model designed specifically for robot operations.
🔍 MiniCPM-RobotTrack: A compact model for real-world object tracking, enhancing robot perception capabilities.
⚙️ PhyAI: A high-performance inference framework that provides efficient support for embodied models, promoting the development of robot understanding, memory, and action capabilities.
More Details: https://github.com/OpenBMB/MiniCPM-Robot
4. DeepSeek V4 Official Version Testing Exposed, Possibly Launching Next Monday Targeting Kimi K3
The test video and real-device demonstration of the DeepSeek V4 official version have been exposed, with impressive performance that may challenge current top models in terms of price and performance.
【AiBase Highlights:】
🧠 DeepSeek V4 shows impressive performance in key metrics, approaching Claude Opus4.8 in real testing.
🎮 In specific tests, the official version of V4 has surpassed Fable5 and GPT-5.6, with significant cost-effectiveness advantages.
💡 Improved interaction experience, with thinking chain style closer to Claude, making output more concise and readable.
5. Alibaba Cloud's BaiLian Launches HappyOyster1.0: An Open-World Model Supporting Real-Time Interaction Begins Grey Testing
Alibaba Cloud's BaiLian has launched HappyOyster1.0, an open-world model supporting real-time interaction, capable of building and exploring AI digital worlds. Unlike traditional text-to-video models, its core training objective is to learn world state transition rules, offering two modes: world exploration and real-time direction.
【AiBase Highlights:】
🧠 World Exploration Mode allows users to input text or upload images to generate interactive digital worlds.
🎬 Real-Time Director Mode enables users to pause, rewind, and adjust the story's direction at any point in the plot.
🌐 HappyOyster1.0 can be applied to open-world game prototypes, interactive short dramas, and virtual companions.
6. Breaking the 15-Second Magic: ZhiXiang Future Launches the World's First Unlimited-Length Creative Agent vivago R1, Commercial Availability Rate Reaches 85%
ZhiXiang Future announced the release of vivago R1 at the 2026 World Artificial Intelligence Conference. This is the world's first unlimited-length content creation multimodal agent, capable of solving issues such as time limitations, logical confusion, and unstable effects in traditional AI video creation. At the same time, ZhiXiang Future has also partnered with multiple institutions to promote the application and ecosystem development of AI technology globally.
【AiBase Highlights:】
🧠 vivago R1 is the world's first multimodal creative agent with the ability to generate and edit videos of unlimited length.
🌐 ZhiXiang Future has formed the Belt and Road Token Export Alliance with institutions like Zhongke Leinao, promoting Chinese AI capabilities globally.
⚙️ ZhiXiang Future launched the AgentOS intelligent agent operating system, supporting multi-agent collaboration and improving industrial implementation efficiency.
7. Kunlun Wanzhi Has Set 2026 as the Year of World Models: Matrix-Game 3.5 Generates in Real-Time on a Single Card, 5B Model Runs at 20FPS
Kunlun Wanzhi announced at the 2026 World Artificial Intelligence Conference that 2026 would be set as the Year of World Models and released Matrix-Game 3.5, Mureka v9.5, and O3 Music Model. These products demonstrate strong real-time interaction capabilities and music creation toolization, marking the advancement of AI technology toward more complex and practical directions.
【AiBase Highlights:】
🧠 Matrix-Game 3.5 world model has patch-level memory injection capability, enabling real-time interactive generation.
🎵 Mureka v9.5 and O3 Music Model support versioned creation, enhancing the tool feel.
🚀 Kunlun Wanzhi has defined 2026 as the Year of World Models, marking the starting point for AI generating interactive worlds.
More Details: https://github.com/Riemann-Dynamics/Matrix-Game-3.5
8. Tencent Huan Yuan Hy3 Free Trial Extension to August 5th, WorkBuddy and CodeBuddy Users Continue to Use Free
Tencent Huan Yuan Hy3 free trial has been extended to August 5th, aiming to further expand its usage scale in real business environments and collect user feedback to optimize model capabilities.
【AiBase Highlights:】
🧠 Tencent Huan Yuan Hy3 free trial has been extended to August 5th, allowing WorkBuddy and CodeBuddy users to continue using it for free.
💻 Hy3 model performs well in code generation, development assistance, and workflow scenarios, attracting many users for testing.
📈 By extending the free trial period, Tencent hopes to expand the usage scale of Hy3 in real business environments and collect user feedback to optimize model capabilities.
