Microsoft Azure AI Adds 40 New Large Models, Covering Text, Image, Code, and Speech Generation


The domestic large model MiniMax M2.5 has ranked first in global API call volume for five consecutive weeks, showcasing China's AI transition from a follower to a setter. Its core advantage lies in extreme cost-effectiveness, achieving the same level of intelligence at a lower cost.
China's AI large models are rapidly rising in the global consumer market, surpassing the U.S. in API calls for two consecutive weeks, securing the top three spots globally. According to OpenRouter data, as of March 15, the weekly API call volume reached 4.69 trillion Tokens. The growth momentum mainly comes from AI Agent applications like OpenClaw, driving a transformation from question-answering to intelligent agents.
The Alibaba Qianwen family launched its flagship model, Qwen3.5-Max-Preview, which made its debut on the global large model competition platform LMArena with a total score of 1464 points, elevating Alibaba Qianwen to the fifth position in the global large model company rankings, firmly securing the top position in China. This marks that domestic large models have entered the top tier in global competition.
Kimi Company released a paper titled "Attention Residuals: Rethinking Depth-Wise Aggregation", proposing a new method of attention residuals to optimize the depth-wise aggregation mechanism. Elon Musk, CEO of Tesla, liked the research on social media, calling it an outstanding work. Kimi official humorously responded, sparking a global discussion in the AI community.
iFLYTEK launches the enterprise-level intelligent assistant AstronClaw, focusing on secure collaboration. It ensures data privacy through sandbox isolation technology, supports cloud-based one-click deployment, and integrates with office platforms such as Enterprise WeChat and DingTalk, helping enterprises efficiently apply large models.