- The new model completed over a thousand tool calls during a 35-hour continuous test, demonstrating unprecedented enterprise-grade stability.
- The release follows the company's recent open-source initiatives, further consolidating Alibaba's leading position in the foundational large model sector.
Alibaba has officially released Qwen3.7-Max, its most powerful agent foundation model to date, aiming to become an all-around agent base capable of autonomously and continuously executing long-cycle tasks.
The new generation flagship model is designed to address the industry challenge of AI (artificial intelligence) agents easily crashing in real-world production environments.
The product's official debut coincided with the Alibaba Cloud summit held on May 20. Previously, preview models bearing this version number had been launched in advance on multiple testing platforms, attracting widespread market attention.
The core breakthrough of Qwen3.7-Max lies in its outstanding stability in handling long-cycle autonomous tasks, according to an announcement on Wednesday.
In an extreme stress test, the model ran autonomously and continuously for about 35 hours on a previously unseen hardware platform, according to the announcement.
During this period, it spanned 1,158 tool calls and achieved zero interruptions. The model completely autonomously completed the performance analysis and refactoring of the underlying architecture, ultimately achieving a 10-fold geometric mean speedup across multiple workloads.
This long-range strategic coherence benefits from the research and development team's thorough reconstruction of environmental diversity during the training phase. They adopted an orthogonal decoupling design, enabling the model to break free from over-reliance on specific execution frameworks.
This means Qwen3.7-Max possesses exceptionally strong cross-framework adaptability. Whether developers use Claude Code, OpenClaw, or other frameworks, the model can achieve plug-and-play functionality while maintaining consistent performance, the announcement said.
In real-world application scenarios, the new model comprehensively reshapes highly complex, enterprise-grade professional workflows. It can not only handle programming tasks ranging from front-end prototypes to complex software engineering, but also achieve office automation through MCP integration.
Benchmark data further confirms the model's all-around performance across multiple tasks. Qwen3.7-Max ranked first among its domestic peers in various evaluations, including coding agents and complex reasoning, with performance approaching top global levels, according to the announcement.
The new model will soon provide API services to the public through Alibaba Cloud's Bailian platform.
This release continues the rapid iteration pace of the Qwen large models, following the company's launch of the highly efficient open-source model Qwen3.6-35B-A3B last month.
