Key Info
Tencent introduces WebCraftBench, a benchmark that evaluates AI-generated web apps using live testing and coverage-guided exploration, matching human preference 85.3% of the time.
Highlights
- Agents use the live app and coverage-guided exploration to find unreached functionality.
- Scores aesthetics, usability, and whether the original request was met.
- Validated on 197 human-validated pairs with 85.3% agreement.
- Paper available at the provided link.