DesignBench: A Multi-framework Multi-task Benchmark for Front-end Code Generation
Code and dataset for the paper.
May 31, 2025
DesignBench: A Comprehensive Benchmark for MLLM-based Front-end Code Generation
We introduce DesignBench, the first comprehensive multi-framework multi-task benchmark for front-end code generation, encompassing React, Vue, and Angular frameworks. Beyond traditional design generation, we pioneer design edit and repair tasks. Through extensive experiments across task complexity, framework compatibility, difficulty levels, contextual factors and in-depth code-level analysis, we reveal primary limitations in current MLLMs for framework-based development and elicit several actionable advice for researchers and developers.
May 30, 2025