Daily AI News Brief - July 29, 2025

July 29, 2025 presents six significant AI developments spanning next-generation reasoning models, revolutionary video generation architecture, AI research assistance, intelligent browsing enhancement, platform usage management, and massive-scale language model releases.

1️⃣ Zhipu AI Major Release GLM-4.5: World's First Open-Source Model Natively Integrating Reasoning, Coding, and Agent Capabilities

Zhipu AI has released next-generation flagship model GLM-4.5, achieving open-source SOTA levels in reasoning, coding, and agent capabilities while demonstrating excellent performance in global evaluations. The model adopts MoE architecture providing high parameter efficiency with full-stack development capabilities and ecosystem compatibility, serving as a powerful technical foundation for agent applications.

Revolutionary Integration Features:Native Capability Fusion: GLM-4.5 first achieves native integration of reasoning, coding, and agent capabilities, meeting complex requirements for diverse AI applications
MoE Architecture Excellence: Adopts Mixture of Experts architecture providing two versions with high parameter efficiency and superior performance optimization
Cost-Effective API: API calling prices far below mainstream models, supporting low-latency, high-concurrency actual deployment needs for enterprise applications

Available on Zhipu AI Open Platform with open-source versions on Hugging Face and ModelScope, GLM-4.5 features 355B total parameters with 32B active parameters, while GLM-4.5-Air offers 106B total with 12B active parameters, both utilizing efficient MoE design for optimal performance.

2️⃣ Alibaba Open Sources Tongyi Wanxiang Wan2.2: World's First MoE Architecture Video Generation Model

Alibaba's open-source Tongyi Wanxiang Wan2.2 achieves major breakthroughs in video generation technology, introducing the industry's first MoE architecture that improves computational efficiency by 50% while featuring cinematic aesthetic control systems. The release provides consumer-grade GPU deployable small-scale models, significantly lowering technical barriers for AI video generation.

Breakthrough Video Generation Features:

MoE Architecture Innovation: Introduces MoE architecture improving computational efficiency by 50% through specialized expert model coordination
Cinematic Aesthetic Control: First-ever cinematic aesthetic control system supporting professional film effects including lighting, color, and composition control
Consumer Hardware Support: Provides 5B unified model enabling consumer-grade GPU deployment for democratized access to professional video generation

Available on GitHub at https://github.com/Wan-Video/Wan2.2, Wan2.2 incorporates 27B total parameters with 14B active parameters, supporting both text-to-video and image-to-video generation at 720P resolution with 24fps performance on consumer graphics cards like RTX 4090.

3️⃣ StepFun Launches AI Research Assistant JieYue ShenYan with Invitation Testing

StepFun AI has launched the new research assistant JieYue ShenYan, demonstrating excellent performance in finance, consulting, healthcare, and other fields while efficiently completing complex research tasks and generating professional research reports. Its powerful information processing capabilities enable outstanding performance across multiple benchmark tests with support for various format exports.

Advanced Research Capabilities:

Domain Expertise: JieYue ShenYan designed specifically for finance, consulting, healthcare fields, completing complex research tasks within ten minutes
Benchmark Excellence: Outstanding performance in xbench-DeepSearch and BrowseComp benchmark tests with industry-leading pass rates
Comprehensive Output: Supports multiple format research report exports with verification capabilities ensuring data authenticity and reliability

The AI research assistant represents advancement in automated research methodology, enabling rapid analysis and synthesis of complex information across professional domains while maintaining high accuracy standards for reliable research outcomes.

4️⃣ Microsoft Edge Browser Launches Copilot Mode for Enhanced Intelligent Search Experience

Microsoft Edge browser has launched the new Copilot mode, utilizing artificial intelligence technology to help users search, organize, and make decisions more efficiently while improving overall work productivity. The mode supports voice navigation with emphasis on privacy protection, allowing users to control data usage at any time.

Intelligent Browsing Features:

Automated Search Integration: Microsoft Edge launches Copilot mode with automated search and information integration, enhancing browsing efficiency
Natural Voice Navigation: Supports natural voice navigation enabling users to interact directly with Copilot through speech commands
Privacy-First Design: Privacy protection as top priority with user data used only for experience improvement, ensuring information security

Available to all users for free with continuous iterative updates, Copilot mode integrates seamlessly into Edge browser providing contextual assistance, video summarization, and intelligent content analysis while maintaining user control over privacy settings and data usage preferences.

5️⃣ Anthropic Implements Usage Restrictions on Claude Users, Causing Developer Dissatisfaction

Anthropic has announced weekly usage limits for Claude users, primarily affecting approximately 5% of users. This move has generated dissatisfaction among some developers and users who view it as collective punishment of compliant users, while users can purchase additional usage quotas to continue their work.

Platform Restriction Details:

Weekly Usage Limits: Anthropic implements weekly usage restrictions on Claude users, primarily affecting approximately 5% of user base
Developer Concerns: Developers express dissatisfaction with throttling policies, considering punishment of compliant users unreasonable
Additional Quota Options: Users can purchase additional usage quotas at standard API rates to ensure project continuity

The restrictions began August 28, 2025, targeting overuse, continuous sessions, and account sharing that affects overall system reliability, with Anthropic stating most subscribers will not be impacted while maintaining platform governance and fair access principles.

6️⃣ StepFun Launches New Foundational Large Model Step 3 with 321B Parameters

StepFun has launched next-generation foundational large model Step 3 with up to 321B parameters, featuring strong intelligence, low cost, open-source availability, and multimodal capabilities. The model demonstrates excellent reasoning efficiency performance, achieving high-efficiency operation across various hardware platforms with planned open-source release on July 31st.

Foundational Model Specifications:

Massive Parameter Scale: Step 3 features up to 321B parameters representing significant advancement in model scale and capability
Multi-Platform Efficiency: Achieves high-efficiency operation across various hardware platforms with optimized inference performance
Open Source Commitment: Planned open-source release on July 31st promoting community development and research advancement

Additionally, StepFun has partnered with multiple chip and platform manufacturers to establish the Model-Chip Ecosystem Innovation Alliance, accelerating large model application implementation across diverse hardware environments and promoting broader AI technology adoption through collaborative ecosystem development.