AI News

Beijing Tiantan Hospital and Yinghe Yimai jointly released 'Dr. Xiaojun 2.0', the world's first full-disease-coverage CT-assisted report generation model for cranial scans. Leveraging Tiantan's massive cranial CT data and Yinghe's foundation model with AI Agent technology, it automates the entire process from image analysis to diagnostic reporting, significantly enhancing neuroimaging diagnostic standards.....

Fuke AI secures strategic investment from Alibaba to enhance e-commerce digital productivity, integrating AI into merchant operations for improved customer service, experience, and operational quality. The collaboration strengthens Fuke AI's product system across service, shopping guidance, products, transactions, and after-sales, accelerating AI alignment with merchant needs.....

Yinghe Yimei and Beijing Tiantan Hospital jointly launched the world's first large model for full-disease coverage in cranial CT auxiliary report generation, "Xiao Jun Doctor 2.0", on April 24 in Beijing. This AI product aims to improve the efficiency and accuracy of medical imaging reports through advanced technology, attracting widespread attention from medical professionals and tech enthusiasts.

The rapid development of AI triggers user anxiety, with Anthropic's Claude Code product lead Cat Wu noting that high-speed iteration causes FOMO, forcing users to constantly check social platforms for updates. Previously, features were released monthly or quarterly, allowing occasional checks, but the faster pace now amplifies unease.....

Tencent Cloud has launched the DeepSeek-V4 preview version API on the TokenHub platform, supporting a million-level context length and significantly enhancing natural language processing capabilities. The pricing is consistent with DeepSeek's official pricing, offering a high-cost-performance service, and is also available on the Singapore node of the international site, covering global users.

Canadian startup Cohere and German startup Aleph Alpha have formed a $20 billion partnership to develop a 'sovereign' AI system, aiming to create an AI architecture independent of the US and China, advancing transatlantic technological autonomy. Cohere specializes in natural language processing, while Aleph Alpha excels in reasoning models; together, they will combine their technological strengths to accelerate independent AI development.....

At the 2026 Beijing Auto Show, BMW partnered with Alibaba Cloud to launch three AI assistants for China: 'Car Expert', 'Travel Buddy', and 'Encyclopedia Master'. These shift voice commands from passive response to active decision-making, enabling 'understand and execute' for enhanced driving and travel. The new BMW iX3 long-wheelbase version debuts with them.....

At the Beijing Auto Show on April 24, Alipay and Banma Intelligence jointly launched a new AI cockpit solution integrating the in-vehicle version of 'AI Pay', upgrading smart cockpits from 'listen and speak' to 'pay and act'. The initial phase covers entertainment and travel scenarios, enabling drivers to purchase tickets, book hotels, and order meals via voice commands.....

Google DeepMind introduces the decoupled DiLoCo distributed training architecture, which enhances training efficiency for large-scale AI models by distributing the process across multiple asynchronous, fault-isolated 'compute islands,' improving robustness against hardware failures and overcoming the limitations of traditional tight synchronization prone to single points of failure.....

In the Hong Kong stock market, shares of Zhipu Technology and Minimax fell significantly after the release of Deepseek V4, a highly anticipated deep learning model with technical upgrades and enhanced features. This unexpected downturn in these major AI concept stocks sparked widespread investor discussion.....

Perplexity CEO Aravind Srinivas argues AI won't disrupt smartphones but will evolve iPhones into 'digital passports'. As AI relies on context, iPhones storing personal data (payments, health, communication) become critical infrastructure, with Apple's chips as an underestimated advantage.....

Cambricon announced the completion of Day0 compatibility for the entire series of DeepSeek-V4 models. Based on the vLLM inference framework, it covers the 285B parameter Flash version and the 1.6T parameter Pro version. By optimizing sparse attention and compressed structures with its self-developed Torch-MLU-Ops operator library, the model can run stably on Cambricon hardware on the day of release. The related code has been open-sourced to GitHub.

The gaming industry is widely adopting AI, particularly generative AI. Ubisoft requires relevant experience from all applicants, while the developer of ARC Raiders has completely overhauled its development process with AI. Although some AAA titles remain cautious, AI has become an industry trend.....

Cambricon announced successful Day 0 adaptation of DeepSeek-V4, an open-source AI model by DeepSeek, achieving stable operation on launch day. Using its self-developed fusion operator library Torch-MLU-Ops, it accelerated modules like Compressor and mHC, significantly boosting inference efficiency. The vLLM inference framework was also adopted for a more efficient AI experience.....

Meituan has recently launched a trillion-parameter AI large model test. The model is trained entirely on domestic computing power clusters, marking a significant breakthrough in the application of domestic technology. It is currently only available to invited users and has not been widely released, demonstrating Meituan's leading position in the AI field.

DeepSeek launched the next-generation model DeepSeek-V4 on April 24, achieving performance comparable to top-tier closed-source models, marking a milestone in open-source AI. It supports a 1-million-token ultra-long context, suitable for long-document analysis, code generation, and multi-turn tasks. Lightweight Flash and standard versions were also released for rapid iteration and deep application needs.....

According to Kunlun Tech's 2025 annual report, the company's revenue reached 8.198 billion yuan, an increase of 44.78% year-on-year, with overseas revenue reaching 7.723 billion yuan, up 49.91%. The company introduced the "4+3 Strategy", clearly defining the development direction of AI-driven content production, covering both technological and business layout.

AI programming company Cursor faced obstacles in its quest for several billion dollars in funding, with its $5 billion valuation deterring several later-stage investment firms. Previously, SpaceX had shown interest in acquiring it for $6 billion, but top funds including Iconiq have clearly rejected the offer. The main reason for the cold funding climate is that global capital has already completed its initial setup in the AI sector.

DeepSeek launches DeepSeek-V4 preview series as open-source, achieving million-character ultra-long context and leading domestic Agent collaboration and knowledge reasoning. Offers Pro (1.6T parameters, 49B activated) and Flash versions; Pro matches top closed-source models, Flash balances efficiency.....

DeepSeek launched its new flagship model V4 with a segmented strategy, introducing Flash and Pro versions for lightweight high-frequency applications and complex reasoning tasks, respectively. This move meets diverse scenario needs and reshapes AI commercialization benchmarks with competitive pricing, integrating the original deepseek-chat and deepseek-reasoner models.....