Alibaba Launches Qwen3.7‑Max and Full‑Stack AI Infrastructure at Cloud Summit

Benchmark reports cited by Chinese media indicate strong performance across coding‑agent evaluations such as SWE‑Pro and SWE‑Multilingual, along with a 69.7 score on Terminal Bench 2.0‑Terminus, placing it ahead of several competing models in those tests.

The model is also designed for workflows that involve:

• complex reasoning across multiple steps
• external tool usage
• integrations such as MCP (Model Context Protocol)
• multi‑agent collaboration for automation and productivity tasks

These capabilities reflect a shift toward AI systems that act as autonomous collaborators rather than simple chat assistants.

Long‑horizon autonomous tasks

Alibaba emphasized Qwen3.7‑Max’s ability to operate for long periods without human intervention.

In one demonstration experiment described at the summit, the model reportedly worked autonomously for about 35 hours, executing more than 1,000 tool calls while maintaining coherent reasoning throughout the process.

During that test, the model optimized a production‑grade AI inference kernel on a new chip platform. The result reportedly delivered around a 10× speed improvement compared with the original implementation.

Such long‑duration autonomy is central to Alibaba’s vision for AI agents capable of executing hundreds or thousands of sequential steps across real‑world workflows.

Benchmark positioning among global models

Reports from third‑party benchmark rankings suggest that Qwen3.7‑Max performs competitively with top global models.

Chinese media coverage citing the Arena global blind‑test leaderboard said the model ranked first among Chinese domestic models, with performance approaching leading systems from OpenAI, Anthropic, and Google.

These benchmark results are based on reported evaluations and should be interpreted cautiously until independently replicated.

New AI chips and infrastructure

The summit announcements went beyond the model itself. Alibaba also revealed major updates to its AI infrastructure, including the Zhenwu M890 AI chip developed by its semiconductor unit T‑Head.

Key details reported at the event include:

• The M890 delivers about three times the performance of the previous Zhenwu 810E chip.
• It includes 144 GB of on‑chip memory designed for high‑throughput AI workloads.
• Alibaba introduced a 128‑card super‑node server architecture built around the new hardware.

The company said this infrastructure will be made available to enterprise customers through Alibaba Cloud’s Bailian platform, supporting large‑scale training and inference workloads.

The strategy: building China’s “AI factory”

Alibaba executives framed these announcements as part of a long‑term strategy to build a vertically integrated AI ecosystem.

At the summit, Alibaba Cloud senior vice‑president Liu Weiguang described AI development as a form of industrial production powered by “training and inference factories.” He said the company aims to create “China’s AI factory” by controlling every layer of the technology stack.

According to Alibaba, that stack includes five major layers:

• custom AI chips
• cloud computing infrastructure
• foundation models such as Qwen
• model service platforms and developer tools
• AI applications and agents

By building capabilities across all these layers, Alibaba aims to become one of the most vertically integrated AI providers in China—similar to how some Western tech companies combine hardware, cloud services, and AI models into unified ecosystems.

Why the announcement matters

The launch of Qwen3.7‑Max is not just another model release. Instead, it highlights a broader industry shift toward agent‑centric AI systems capable of executing long chains of reasoning and real‑world tasks autonomously.

For Alibaba, the model serves as the software centerpiece of a much larger platform strategy that combines:

• proprietary chips
• hyperscale cloud infrastructure
• foundation models
• developer platforms
• downstream AI applications

Together, these components form the backbone of the company’s effort to industrialize AI development—what it describes as building a national‑scale AI production system.

If successful, that approach could position Alibaba as a major global competitor in the race to deliver full‑stack AI infrastructure for the agentic computing era.