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://top.aibase.com/
1. Zhipu AI Launches GLM-4.5: The World's First Native Integrated Reasoning, Coding, and Agent Capability Open Source Model
Zhipu AI has released its new flagship model, GLM-4.5, which achieves state-of-the-art performance in reasoning, code, and agent capabilities globally. It adopts a MoE architecture, offering high parameter efficiency and full-stack development capabilities with ecological compatibility, providing a strong technical foundation for agent applications.
【AiBase Summary:】
🧠 GLM-4.5 is the first to achieve native integration of reasoning, coding, and agent capabilities, meeting complex needs.
⚙️ Adopting a mixture of experts (MoE) architecture, it offers two versions with high parameter efficiency and superior performance.
💰 API call prices are far lower than mainstream models, supporting low-latency, high-concurrency practical deployment needs.
2. Alibaba Opens Source Tongyi Wan 2.2: The World's First MoE Architecture Video Generation Model Shockingly Released
Alibaba's open-source Tongyi Wan 2.2 has made a major breakthrough in video generation technology, introducing the industry's first MoE architecture, improving computational efficiency, and featuring a cinematic aesthetics control system. It also provides a small-sized model that can be deployed on consumer-grade graphics cards, significantly lowering the technical barriers for AI video generation.
【AiBase Summary:】
✨ Introduces MoE architecture, improving computational efficiency by 50%
🎬 First-ever cinematic aesthetics control system, supporting professional film effects such as lighting and color.
🚀 Provides a 5B unified model, deployable on consumer-grade graphics cards.
Details link: https://github.com/Wan-Video/Wan2.2 HuggingFace: https://huggingface.co/Wan-AI
3. Step Star Unveils AI Research Assistant Step Deep Research and Opens Beta Testing
Step AI launched a new research assistant called 'Step Deep Research,' which excels in finance, consulting, and healthcare fields. It can efficiently complete complex research tasks and generate professional research reports. Its powerful information processing capability makes it perform well in multiple benchmark tests and supports exporting in various formats.
【AiBase Summary:】
✨ Step Deep Research is designed for finance, consulting, and healthcare fields, capable of completing complex research tasks within ten minutes.
📊 Performs well in the xbench-DeepSearch and BrowseComp benchmark tests, with a leading pass rate in the industry.
📄 Supports exporting research reports in multiple formats and has verification capabilities to ensure data authenticity and reliability.
4. Microsoft Edge Browser Launches Co-pilot Mode, Enhancing Intelligent Search Experience
Microsoft Edge browser launched a new Co-pilot mode, helping users search, organize, and make decisions more efficiently using artificial intelligence technology, improving overall work efficiency. This mode supports voice navigation and emphasizes privacy protection, allowing users to control data usage at any time. Currently, the Co-pilot mode is available for free to all users and will continue to be iterated and updated.
【AiBase Summary:】
🧑💻 Microsoft Edge launches Co-pilot mode, automating search and information integration to improve browsing efficiency.
🗣️ Supports natural voice navigation, allowing users to interact directly with Co-pilot through voice.
🔒 Privacy protection is a top priority, with user data used only to improve experience, ensuring information security.
5. Anthropic Implements Usage Limits on Claude Users, Causing Developer Frustration
Anthropic announced weekly usage limits for Claude users, mainly affecting about 5% of users. This move caused dissatisfaction among some developers and users, who believe it is an unfair punishment for compliant users. Meanwhile, users can purchase additional usage quotas to continue working.
【AiBase Summary:】
🌟 Anthropic will implement weekly usage limits on Claude users, mainly affecting about 5% of users.
🔧 Developers express dissatisfaction with the usage limit policy, considering it unreasonable to penalize compliant users.
💡 Users can purchase additional usage quotas at standard API rates to ensure project continuity.
6. Step Star Launches New Foundation Large Model Step 3 with 321B Parameters
Step Star launched a new foundation large model Step 3 with 321B parameters, featuring strong intelligence, low cost, open source, and multimodal capabilities. The model shows excellent performance in reasoning efficiency, capable of running efficiently on various hardware platforms. It plans to open source on July 31st. In addition, Step Star has formed the "Modular Chip Ecosystem Innovation Alliance" with multiple chip and platform manufacturers to accelerate the application of large models.
【AiBase Summary:】
🧠 Step 3 is a new foundation large model launched by Step Star, featuring strong intelligence, low cost, open source, and multimodal capabilities.
🚀 Step 3 shows excellent performance in reasoning efficiency, capable of running efficiently on various hardware platforms.
🤝 Step Star has established the "Modular Chip Ecosystem Innovation Alliance" with multiple chip and platform manufacturers to accelerate the application of large models.
7. Ant Group's Data Science Department Launches Financial Reasoning Large Model Agentar-Fin-R1
Ant Group's Data Science Department launched the financial reasoning large model Agentar-Fin-R1 at the World Artificial Intelligence Conference. Based on Qwen3, this model focuses on the professionalism and security of the financial industry, demonstrating outstanding performance.
【AiBase Summary:】
🧠 Agentar-Fin-R1 is a financial reasoning large model based on Qwen3, focusing on professionalism, reasoning ability, and security compliance in the financial industry.
📊 It performs well on multiple authoritative evaluation benchmarks, surpassing open-source general large models of the same size. It achieved excellent results in financial large model evaluations such as FinEval1.0 and FinanceIQ.
🏦 Offers multiple parameter versions to meet the needs of different financial institutions, while building a comprehensive financial task classification system covering banking, securities, insurance, and more.
Details link: https://arxiv.org/pdf/2507.16802
8. Shanghai AI Lab Opens Source Multimodal Large Model "ShuSheng" Intern-S1
Shanghai AI Lab released the multimodal large model Intern-S1, aiming to enhance the depth and breadth of scientific research and showing excellent performance in multiple fields.
【AiBase Summary:】
🧪 Intern-S1 outperforms existing top closed-source models in fields such as chemistry, materials science, and earth sciences.
🧠 The model features a cross-modal scientific analysis engine, accurately interpreting various complex scientific data.
🚀 Shanghai AI Lab will continue to promote the open source and application of Intern-S1, supporting researchers to explore new paths for scientific discovery.
Details link: https://chat.intern-ai.org.cn/
9. Doubao App Upgrades Visual Reasoning Capabilities, Supporting Deep Thinking in Image Analysis
Doubao App has made significant upgrades in visual reasoning, supporting a deep thinking mode for image analysis, providing a smarter experience. Users can upload images for detailed processing and searching, improving search accuracy. At the same time, Doubao can compare historical archives based on detail information to determine the age range of the image and judge the specific location and city name through terrain, architectural styles, and other characteristics.
【AiBase Summary:】
📷 Supports a deep thinking mode, enhancing image analysis accuracy.
🔍 Enables thinking and searching simultaneously, improving search result accuracy.
🗺️ Analyzes terrain, architectural styles, and other features to accurately locate cities.
10. Cyberspace Administration of China Cracks Down on False Information from Self-Media
The Central Cyberspace Administration announced a two-month special campaign to crack down on self-media publishing false information, aiming to solve problems such as malicious hype, data fraud, and qualification fraud through technical governance and platform responsibility.
【AiBase Summary:】
🧠 Malicious hype behavior includes impersonating individuals involved in hot events or insiders to fabricate false information.
📊 Data fraud methods include using AI synthesis technology to forge news or creating fake events through editing and compiling.
👩🔬 Misinformation in specialized fields, where unverified or fake qualifications are used to impersonate experts and publish statements against common sense.