Alibaba's Qwen3.7-Max runs 35 hours autonomously to optimize chip code

Alibaba's Qwen team has unveiled Qwen3.7-Max, a new proprietary AI model designed for autonomous task execution. In a demonstrative test, the model ran continuously for 35 hours to optimize code for its own custom chip without human intervention. This showcases the model's ability to handle complex software projects over extended periods, maintaining performance and consistency across different agent frameworks. The model represents a significant leap in self-directed AI, where it can not only understand code but also iteratively improve it for hardware-level optimizations.

Qwen3.7-Max's 35-hour autonomous run highlights a key capability: long-horizon reasoning and execution. Unlike many models that require frequent human feedback, this model chained together thousands of steps—from code analysis, to rewriting, to testing—all without a single manual correction. It performed consistently across multiple agent frameworks, indicating robust integration flexibility. For enterprises, this means potential for fully autonomous software maintenance, chip optimization, and large-scale code refactoring, reducing the need for human engineers in routine or complex iterative tasks. The release positions Alibaba as a serious competitor in the autonomous AI space, challenging models like OpenAI's o1 and Anthropic's Claude.