Alibaba and others release coding agent benchmark Co-Design Bench
Understand the story
On November 8, 2026, The Decoder reported that researchers from Alibaba and several Asian universities introduced Co-Design Bench, a benchmark for coding agents. The benchmark is built from real users' change requests, requiring agents to modify existing web applications based on various input forms such as text, images, audio, video, or documents. Evaluation results show that among the ten models tested, even the best-performing model could fully complete only about a quarter of the tasks, indicating that current coding agents remain fairly limited in handling such multimodal, real-world application modification tasks.
AI overview of reporting · Updated 2 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderAI coding agents still need human cleanup on nearly three out of four tasks in a real-world test
Researchers from Alibaba and several Asian universities built Co-Design Bench, a coding agent benchmark based on real user change requests, requiring agents to modify existing web applications based on inputs such as text, images, audio, video or documents. Even the best of the ten models tested could fully complete only about a quarter of the tasks.
Story heat trend
Not enough continuous observations to show a trend yet.