Welcome to the 【AI Daily】column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with the hottest AI news, focusing on developers and helping you understand technology trends and innovative AI product applications.
Check out the latest AI products Learn More: https://top.aibase.com/
1. Zhipu Releases CogView4, the First Open-Source Text-to-Image Model Capable of Generating Chinese Characters
On March 4, 2025, Beijing Zhipu Huazhang Technology Co., Ltd. launched CogView4, the first open-source text-to-image model capable of generating Chinese characters. It boasts powerful bilingual input and image generation capabilities. The model ranked first in the DPG-Bench benchmark test, becoming the SOTA among open-source text-to-image models. It supports arbitrary resolution and complex semantic alignment, meeting the needs of advertising and short video industries.
【AiBase Summary:】
🎨 CogView4 is the first open-source text-to-image model capable of generating Chinese characters, with strong bilingual input capabilities.
🚀 It ranked first in the DPG-Bench benchmark test, becoming the SOTA among open-source text-to-image models.
🔧 CogView4 supports the Apache2.0 license and will add ecosystem support and fine-tuning toolkits in the future.
Details: https://github.com/THUDM/CogView4
2. Major Breakthrough! MiniMax Launches New Image Generation Model Image-01 at 1/10 the Cost
MiniMax recently launched its first text-to-image generation model, Image-01, marking a new advancement in AI image generation technology. The model is notable for its excellent prompt control and visual composition capabilities, allowing users to create logically consistent and artistic images. Furthermore, Image-01's efficient batch processing and competitive pricing make professional-level AI creation more accessible. Users can experience movie-quality images at a lower cost, greatly expanding their creative space.
【AiBase Summary:】
🌟 Image-01 is a new text-to-image generation model designed to expand users' creative space.
🎨 It features precise prompt control and excellent visual composition capabilities, generating logically consistent and artistic images.
💰 Image-01's service cost is only one-tenth that of similar products, making professional-level AI creation more accessible.
Details: https://www.minimax.io/platform/login
3. Tencent Yuanbao Surpasses DeepSeek to Top Apple's Free App Download Chart
Tencent Yuanbao achieved remarkable success on the Apple China App Store's free app download chart, surpassing DeepSeek to become the most downloaded application. Tencent strengthened its promotion and upgraded its technology, increasing Yuanbao's visibility and appeal. The integration with WeChat, in particular, significantly improved the user experience.
【AiBase Summary:】
📊 Tencent Yuanbao surpassed DeepSeek to become the top app on Apple's free download chart.
📱 Tencent increased promotion of Yuanbao, especially on WeChat.
🚀 Technological upgrades and feature optimizations have ensured Yuanbao's rapid growth.
4. Mita AI Search Launches "Video" Search Function
On March 3, 2025, Shanghai Mita Network Technology Co., Ltd. launched a new AI search function, adding a video search module. This function analyzes a large amount of video content to help users more efficiently find the learning and entertainment videos they need. Users can quickly locate videos of interest, enhancing the search experience.
【AiBase Summary:】
🎥 Added video search module, covering multimodal data and improving user experience.
🔍 Users can quickly locate videos of interest, such as courses or advertising analysis.
🚀 Integrated with DeepSeek R1 inference model to further enhance search intelligence.
5. Revolutionizing City Modeling! AI-Generated 3D City Model GaussianCity Achieves 60x Speed Improvement!
A research team from Nanyang Technological University achieved a significant breakthrough in 3D city model generation with their GaussianCity technology. This technology achieves a 60x speed improvement and supports unbounded 3D city generation. It has been accepted by CVPR2025. GaussianCity uses an innovative BEV-Point representation method and a spatially aware Gaussian attribute decoder, greatly improving computational efficiency and scalability.
【AiBase Summary:】
🚀 GaussianCity technology achieves a 60x speed improvement, breaking the scale limitations of traditional methods.
🌆 Uses the BEV-Point representation method, significantly reducing VRAM requirements and enabling unbounded 3D city generation.
📈 This technology is open-source, bringing new application potential to virtual reality, autonomous driving, and urban planning.
Details: https://github.com/hzxie/GaussianCity
6. Security Warning! Ollama Large Model Tool Found to Have Critical Vulnerabilities
The National Cyber Security Coordination Center recently issued a security warning regarding the open-source cross-platform large model tool Ollama, pointing out that it has serious security vulnerabilities under default configuration, potentially leading to unauthorized access and model theft. Experts have analyzed that privately deployed users face greater risks, as attackers can exploit these vulnerabilities to conduct data breaches and service disruptions.
【AiBase Summary:】
🔍 Ollama's security vulnerabilities are serious under default configuration, especially for privately deployed users.
💻 Attackers can exploit unauthorized access to steal model information and disrupt services, causing significant losses.
🛡️ Users need to strengthen their security awareness, promptly check for vulnerabilities, and immediately report any signs of attack to relevant authorities.
7. Powerful Video Generation Model Wan2.1 GP: Create Blockbusters Even with Low-Spec GPUs!
The DeepBeepMeep team's Wan2.1GP is a video generation model optimized for users with low-end GPUs, based on Alibaba's Wan2.1. It aims to provide powerful video generation capabilities for users lacking high-performance GPU resources. The model performs excellently in multiple benchmark tests, supports various generation tasks, and has Chinese and English text generation capabilities, marking a significant advancement in video generation technology.
【AiBase Summary:】
👍 SOTA Performance: Wan2.1GP excels in multiple benchmark tests, surpassing existing open-source and commercial solutions.
🖥️ Strong Compatibility: Requires only 8.19GB of VRAM, supporting almost all consumer-grade GPUs, suitable for low-end users.
📜 Multi-task Support: Supports various generation tasks such as text-to-video and image-to-video, and has Chinese and English text generation capabilities.
Details: https://github.com/deepbeepmeep/Wan2GP
8. Volcano Engine Announces Open-Sourcing of Large Model Applications: Launches "Large Model Application Lab" to Unleash AI Innovation Potential
Volcano Engine announced via the X platform the official open-sourcing of its large model applications and the launch of the new "Large Model Application Lab" platform. This move marks a significant breakthrough in its AI technology ecosystem construction, aiming to accelerate innovation within the developer community. The open-sourced applications include cutting-edge AI tools such as mobile assistants and Deep Research, generating significant user discussion and showcasing Volcano Engine's leading technological capabilities.
【AiBase Summary:】
📱 Open-source applications include mobile assistants and Deep Research, providing various AI tools to promote developer innovation.
🌐 The "Large Model Application Lab" is launched to provide an open experimental environment for developers and researchers to explore the potential of large model applications.
💬 The community response is enthusiastic, with users believing this move lowers the barrier to AI development and promotes industry collaboration and innovation.
9. QQ Browser Launches "AI Essay Tutoring" Function
QQ Browser's "AI Essay Tutoring" function aims to help students improve their writing skills, rather than simply providing answers. The function provides full-process writing skills guidance, assisting students at every step, from ideation and outlining to material gathering and essay generation. The AI tool emphasizes its supporting role, encouraging students to read and think critically to master writing skills instead of simply copying and pasting.
【AiBase Summary:】
🧠 The AI essay tutoring function helps students improve their abilities throughout the writing process, rather than just providing answers.
📚 The function provides full-process guidance, covering every step from ideation, outlining, material gathering, and essay generation.
✍️ QQ Browser emphasizes the importance of reading and thinking critically, encouraging students to master writing skills.
10. Anthropic Secures $3.5 Billion in Funding, Valuation Soars to $61.5 Billion
Anthropic recently completed a $3.5 billion Series E funding round, bringing its post-money valuation to $61.5 billion, reflecting investors' continued enthusiasm for the AI market. The round was led by Lightspeed Venture Partners and attracted numerous well-known investors. The funds will be used to advance the development of next-generation AI systems. Despite fierce market competition, Anthropic remains focused on its B2B strategy, committed to providing technical support to global businesses.
【AiBase Summary:】
💵 Anthropic secured $3.5 billion in Series E funding, reaching a valuation of $61.5 billion, reflecting investors' continued enthusiasm for the AI market.
📈 The company's annual revenue reached $1 billion in 2024, a tenfold year-over-year increase, and increased another 30% at the beginning of 2025.
🤝 Amazon and Google invested $8 billion and $3 billion respectively in Anthropic, supporting its B2B strategy.
11. From Coding to Creative Writing: xAI Grok-3 Defeats GPT4.5 to Top the Large Model Arena
xAI's recently launched Grok-3 model performed exceptionally well on the Chatbot Arena leaderboard, narrowly surpassing GPT4.5-Preview. The model, named "grok-3preview-02-24," demonstrated superior performance in various areas, including coding, mathematics, and creative writing. Although referred to as the latest production model, the term "preview" suggests it's still in the testing phase, potentially prompting discussion about its production readiness.
【AiBase Summary:】
🏆 Grok-3 narrowly surpassed GPT4.5-Preview in Chatbot Arena, receiving over 3,000 votes, demonstrating its exceptional performance.
🧠 The model excelled in challenging prompts, coding, mathematical problems, and creative writing, showcasing its powerful capabilities.
🔍 Chatbot Arena uses the Elo rating system for model ranking, marking a significant advancement for xAI in the AI field.
12. Lei Jun's Two Sessions Proposal: Strengthen Governance of "AI Deepfake" Illegal Infringement Hotspots
At the 2025 National Two Sessions, Xiaomi Group founder Lei Jun proposed five key suggestions to promote the development of autonomous driving, intelligent connected new energy vehicles, and artificial intelligence terminals. He emphasized the importance of improving the artificial intelligence terminal standard system, calling for the establishment of intelligent grading standards oriented towards user experience and strengthening the effective connection between international and domestic standards.
【AiBase Summary:】
🚗 Accelerate the mass production of autonomous driving and improve the efficiency of technology application.
🔋 Develop an intelligent connected new energy vehicle industry ecosystem to promote green travel.
⚖️ Strengthen governance of illegal infringements involving "AI deepfakes" to protect user rights.
