Vision2Web: A Hierarchical Benchmark for Visual Website Development with Agent Verification

Recent advances in large language models have improved the capabilities of coding agents, yet systematic evaluation of complex, end-to-end website development remains limited. To address this gap, we introduce Vision2Web, a hierarchical benchmark for visual website development, spanning from static UI-to-code generation, interactive multi-page frontend reproduction, to long-horizon full-stack website development. The benchmark is constructed from real-world websites and comprises a total of 193 tasks across 16 categories, with 918 prototype images and 1,255 test cases. To support flexible, thorough and reliable evaluation, we propose workflow-based agent verification paradigm based on two complementary components: a GUI agent verifier and a VLM-based judge. We evaluate multiple visual language models instantiated under different coding-agent frameworks, revealing substantial performance gaps at all task levels, with state-of-the-art models still struggling on full-stack development.

Modern AI systems can write code, but building a useful website is more than producing a few lines of code. A website often needs to match a visual design, support buttons and page changes, and sometimes manage data behind the scenes. We wanted to understand how well today’s AI systems can handle this kind of complete website-building work. To study this, we created Vision2Web, a collection of realistic website development tasks based on real websites. These tasks range from recreating a single webpage from a design image, to building interactive multi-page websites, to creating more complete websites with both user-facing pages and background logic. We also built an automatic evaluation process that opens the generated websites in a browser, checks whether key actions work, and compares the result with the intended design. Our evaluation shows that current AI coding systems can handle some simpler webpage tasks, but they still struggle when websites require longer sequences of interactions or more complete functionality. Vision2Web provides a practical way to measure this gap and track future progress toward AI systems that can build reliable, realistic websites from user requirements.