Vision2Web: Hierarchical Benchmark Tests End-to-End Website Development Agents with Automated Verification
arXiv·high signal
Vision2Web introduces a hierarchical evaluation framework for coding agents building complete websites from visual designs — not just generating components but full end-to-end development with automated agent-based verification. The benchmark decomposes website fidelity into structural, stylistic, and functional layers. Directly targets the vibe coding wave where agents generate entire web apps from screenshots or mockups.