Alibaba’s Qwen model has broken into the top tier of a global AI coding leaderboard, putting a Chinese-developed system ahead of deployed models from OpenAI and Google.
Qwen3.7-Max ranked fourth on Code Arena’s latest WebDev leaderboard, trailing several Anthropic Claude models while outperforming listed entries from two of the most closely watched US AI labs. The result gives Alibaba a rare top-five spot in a benchmark that tests how well AI models build web apps, not just answer coding prompts.
The result does not mean Alibaba has overtaken OpenAI or Google across AI overall. It does, however, put coding agents and developer tools more firmly at the center of the global AI race.
Alibaba gains ground in AI coding
Alibaba’s Qwen3.7-Max scored 1,541 and ranked fourth on Code Arena’s latest coding leaderboard, the South China Morning Post reported.
The outlet reported that the model outperformed those from OpenAI and Google. SCMP also noted that Anthropic’s Claude models held the other top-five positions, making Alibaba the only non-US developer in that group.
Alibaba Cloud positioned Qwen3.7-Max as a model built for agent-driven workflows, including coding, office automation, and long-running task execution.
“Qwen3.7-Max is our most versatile and capable model for agent-driven workflows,” the Qwen team wrote.
Code Arena focuses on real app building
According to Tech in Asia, Code Code Arena, formerly known as WebDev Arena, uses blind votes on anonymized model outputs to evaluate how well AI systems build web applications from user prompts.
That approach differs from traditional coding benchmarks such as HumanEval or SWE-bench, which rely on standardized tests. Code Arena is designed to reflect how developers judge model output when building interactive web apps and other product-style software.
The benchmark has expanded beyond simple HTML pages and front-end components to include multi-file React apps, dashboards, browser games, and other product-style software tasks.
Qwen moves into developer tools
Alibaba has connected Qwen’s coding capabilities to Qwen Code, an open-source terminal agent. Tech in Asia emphasized that Qwen Code can access Alibaba Cloud ModelStudio APIs and integrate with development environments, CI/CD workflows, and HTTP services.
Alibaba Cloud also positioned Qwen3.7-Max as a model built for agent-driven workflows, including coding, office automation, and long-running task execution.
Alibaba now has another way to compete for developers as AI coding tools move beyond code completion and into bigger software workflows. The Code Arena result is only one benchmark, and rankings can change quickly, but Qwen3.7-Max’s placement shows that global competition in AI coding is widening beyond OpenAI, Google, and Anthropic.
Also read: Google is rolling out new AI coding tools as it works to challenge Anthropic and OpenAI in software development.