Alibaba's Qwen3.7-Plus automates 10k lines of code, but at what cost?
alibaba released qwen3.7-plus, a multimodal agent that autonomously coded a 10k-line app, signaling a shift from chatbots to systems that control entire operating systems.
alibaba released qwen3.7-plus, a multimodal ai agent built to autonomously operate guis, write code from visual templates, and manage cloud infrastructure. demos showed the system writing 10,000 lines of code over 11 hours and recreating desktop apps like macos stocks. while the model excels at screen-based tasks, outperforming competitors on androidworld and screenspot pro, it lags behind in complex text-based reasoning and scientific benchmarks. it supports anthropic's api protocol, allowing direct integration with claude code, and is available as a proprietary service via alibaba cloud.