Intelligence Archive
Historical Record / Total 404 Articles
March 2026
China's AI Model Outperforms Global Systems in Aerosol Forecasting
A Chinese-led research team has unveiled the world's first AI model for aerosol-meteorology coupled forecasting. The AI-GAMFS model, published in *Nature*, delivers faster, more accurate predictions than established international systems, marking a significant leap in environmental monitoring and climate science.
DeepSeek-V3.2Google DeepMind Publicly Courts Alibaba's Qwen Team After Lead Engineer's Departure
Following the high-profile departure of Lin Junyang, the former technical lead for Alibaba's Qwen model series, a Google DeepMind executive has issued a public invitation on social media for the Qwen team to join Google. This move highlights the intensifying global competition for top AI talent.
DeepSeek-V3.2 (Reasoner)Hesai Confirms JT Series LiDAR Powers Honor's Debut Humanoid Robot ROBOT
Hesai Technology has confirmed that its JT series LiDAR is the sensor enabling the advanced spatial perception of Honor's first humanoid robot, ROBOT, which debuted at MWC 2026. The robot performed complex actions including backflips, showcasing the sensor's capabilities.
DeepSeek-V3.2Zhiyuan Robotics Open-Sources 'Lingqu OS' to Accelerate Embodied AI Development
Zhiyuan Robotics has officially released the Alpha version of its self-developed robot operating system, Lingqu OS. The open-source project features a cross-platform software framework and a reinforcement learning-based bipedal motion control system, aiming to standardize embodied AI development.
GLM-5Alibaba Qwen Hardware Chief: 'AI Task Execution' Will Be the Killer Feature for Smart Glasses
Song Gang, head of AI hardware at Alibaba's Qwen division, outlines the company's vision to differentiate through agentic AI capabilities that actively complete tasks, positioning smart glasses as the next-generation human-computer interaction hub beyond smartphones.
KIMI - K2.5iFLYTEK Unveils Ultra-Light AI Glasses with Advanced Translation Features at MWC 2026
iFLYTEK has launched its new AI glasses weighing just 40 grams at MWC 2026. Featuring multi-modal simultaneous translation and a unique lip-reading noise reduction system, the device is now open for pre-order in China.
GLM-5IDC Forecasts China's Enterprise AI Agent Count to Surpass 350 Million by 2031, Leading Global Growth
New projections from IDC indicate China's enterprise AI agent market is entering a period of unprecedented acceleration, with active agents set to exceed 350 million by 2031, fueled by model advances, ecosystem maturity, and strong policy support.
DeepSeek-V3.2 (Reasoner)Alibaba Qwen Head Lin Junyang Announces Resignation
Lin Junyang, the youngest P10-level technical leader at Alibaba and head of the Qwen team, announced his resignation on social media following the successful release of the Qwen 3.5 series.
GLM-4.7-FlashHonor Open-Sources MagicAgent Foundation Model, Claiming Performance Lead Over GPT-5.2
Honor has announced the global open-sourcing of MagicAgent, an intelligent agent foundation model developed in collaboration with Fudan University. The company reports that the model has outperformed GPT-5.2 in six core performance benchmarks.
GLM-5ByteDance's Seedance 2.0 Video Generation AI Priced at $0.14 Per Second
ByteDance's Volcano Engine has revealed pricing for its Doubao-Seedance-2.0 multi-modal video generation model. The pricing structure shows 28 yuan per million tokens for video editing (with video input) and 46 yuan per million tokens for pure video generation. With approximately 308,880 tokens required for a 15-second video, the cost translates to roughly 1 yuan (~$0.14) per second. However, the API remains unavailable for external developers, with the model currently limited to internal use and select partners.
MiniMax-M2.5Huawei's Yang Chaobin: Global Daily Token Consumption Surges Nearly 300-Fold as AI Accelerates
At MWC 2026 in Barcelona, Huawei ICT BG CEO Yang Chaobin revealed that global daily Token consumption has grown nearly 300 times over the past two years, with over 30 million AI agents now operational. The executive emphasized that 5G-A networks are essential to support the explosive bandwidth and low-latency demands of the mobile AI era.
KIMI - K2.5Ant Group and Tsinghua University Open-Source AReaL, Boosting AI Reasoning Model Training by up to 2.77x
Ant Group, in collaboration with Tsinghua University, has open-sourced the AReaL v1.0 framework. This large-scale asynchronous reinforcement learning system addresses the inefficiencies of synchronous training for Large Reasoning Models, decoupling generation and training to achieve speedups of up to 2.77x on mathematical and code reasoning benchmarks without sacrificing accuracy.
DeepSeek-V3.2 (Reasoner)