HANGZHOU, 23 May 2025 (The Capital Post) — Chinese technology giant Alibaba has announced a major upgrade to its full AI ecosystem, introducing new cloud infrastructure, AI chips, and foundation models aimed at accelerating the development of AI agents worldwide.
The announcement was made during the Alibaba Cloud Summit, where the company introduced its latest large language model, Qwen3.7-Max, designed to support advanced coding, complex reasoning, and long-horizon task execution. The model is expected to be made available soon for developers and enterprises globally.
According to Alibaba, Qwen3.7-Max is built to handle agentic AI tasks such as code generation, workflow automation, and large-scale multi-step operations involving thousands of actions. The company said the model is capable of operating continuously for up to 35 hours while managing over 1,000 tool calls without performance degradation.
The model has also been optimised to work with leading AI agent frameworks including OpenClaw, Hermes Agent, Claude Code, Qwen Paw, and Qoder, positioning it as a versatile solution for enterprise AI deployment.
-Advertisement-
In addition to the new AI model, Alibaba Cloud unveiled the Panjiu AL128 Supernode Server, powered by the latest Zhenwu M890 AI processor and ICN Switch 1.0 networking chip. The server integrates 128 AI accelerators within a single rack and delivers petabyte-per-second scale bandwidth to support large-scale AI inference and training workloads.
Alibaba’s model service platform, Bailian, also introduced Agentic RL, a reinforcement learning mechanism designed to improve model performance through agent execution feedback. The platform includes built-in safety governance features to ensure AI agents operate within controlled boundaries.
Meanwhile, Alibaba subsidiary T-Head introduced the Zhenwu M890 AI accelerator, which reportedly delivers three times the performance of its predecessor. The chip features 144GB GPU memory, 800GB per second inter-chip bandwidth, and supports multiple data precision formats ranging from FP32 to FP4 for both training and inference tasks.
T-Head also launched the ICN Switch 1.0 networking chip and its proprietary T-Head SAIL™ software stack to further improve computational efficiency and large-scale AI processing capabilities. The company revealed that over 560,000 Zhenwu units have been delivered globally, with adoption spanning more than 400 customers across 20 industries. – The Capital Post
