Alibaba Qwen3.7-Max ranks #4 globally on Code Arena WebDev, beating OpenAI and Google

Alibaba's Qwen3.7-Max climbed to fourth globally on Code Arena's WebDev leaderboard on May 28, beating deployed models from OpenAI and Google on web application building benchmarks. The top three remain Anthropic Claude models, making Alibaba the only non-US developer in the top five — a notable marker in the US-China AI competition narrative, particularly given the model's free availability via Alibaba Cloud.

Alibaba positions Qwen3.7-Max as a versatile agentic model spanning coding, office automation, and long-horizon autonomous task execution. The company demonstrated the model running autonomously for extended periods on complex multi-step tasks, which is the bar all frontier labs are now competing on. Separately, the official @Alibaba_Qwen X account announced Qwen3.5 reaching a record 580 tokens-per-second for agentic workloads on the TokenSpeed inference engine, crediting partnerships with Lightseek, NVIDIA AI, the Mooncake team, and Tri Dao's group.

Competitive context: this is the third frontier-class Chinese model launch in two weeks, alongside DeepSeek V4-Pro (now with a permanent 75% price cut, see separate story) and Moonshot/Liquid AI's small-efficient-model moves. Anthropic's Claude Code dominance at the top of the leaderboard remains intact, but the gap below Claude is now occupied by Chinese labs rather than OpenAI and Google. Code Arena is a credible benchmark (LMSys-adjacent crowd-sourced eval), but skeptical takes note that WebDev specifically favors models tuned for verbose multi-file output, which may not generalize to backend or systems code.

Watch next: whether Qwen3.7-Max gets US enterprise adoption given export-control overhang, and whether the next Qwen release closes the remaining gap to Claude. The community read on Hacker News is that the Chinese open-weight ecosystem is becoming a structural threat to closed US frontier labs on price-performance.