Ali Qwen3.7-Max Programming Ranks Fourth Globally, Surpassing OpenAI and Google

According to a report by the South China Morning Post on May 27, Alibaba's Qwen3.7-Max scored 1541 points on the CodeArena leaderboard, surpassing similar products from OpenAI and Google. This is also the first time a Chinese AI model has made it into the top five of the list.

Unlike traditional programming benchmarks such as HumanEval and SWE-bench, CodeArena was founded by the University of California, Berkeley, in collaboration with the University of San Diego and Carnegie Mellon University. It uses real developer blind testing and voting to grade, assessing the model's ability to independently build complete interactive web applications from scratch. This mechanism makes its results more closely aligned with actual development scenarios.

Qwen3.7-Max is specifically designed for autonomous tasks, capable of independently managing long-term workflows, invoking software tools, and writing code on its own. Alibaba disclosed on WeChat that this model can handle complex tasks continuously for up to 35 hours and can continuously invoke tools thousands of times without human intervention.

This release reflects a general shift in China's AI industry. Domestic competitor DeepSeek also recently announced the recruitment of product managers and engineers in the direction of programming proxies. Its senior researcher, Chen Deli, stated on social media that the goal is to develop a programming tool chain "benchmarked by Claude Code."

The programming track has special strategic value for Chinese AI companies. Since software development relies on globally standardized programming languages, the internationalization barrier in this field is much lower than that of consumer-facing internet services. However, Cursor, GitHub Copilot, and Claude Code still dominate the daily work of global developers.

Both Microsoft CEO Satya Nadella and Anthropic CEO Dario Amodei have pointed out that the long-term success or failure of AI competition ultimately depends on who can truly integrate into developers' daily work, not just the temporary scores on the leaderboard.