ArtifactsBench: Bridging the Visual-Interactive Gap in LLM Code Generation Evaluation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918150271926272 |
|---|---|
| author | Zhang, Chenchen Li, Yuhang Xu, Can Liu, Jiaheng Liu, Ao Zhou, Changzhi Deng, Ken Wu, Dengpeng Huang, Guanhua Li, Kejiao Yi, Qi Xiong, Ruibin Hu, Shihui Zhang, Yue Jiang, Yuhao Xu, Zenan Zhang, Yuanxing Zhou, Wiggin Zhou, Chayse Lian, Fengzong |
| author_facet | Zhang, Chenchen Li, Yuhang Xu, Can Liu, Jiaheng Liu, Ao Zhou, Changzhi Deng, Ken Wu, Dengpeng Huang, Guanhua Li, Kejiao Yi, Qi Xiong, Ruibin Hu, Shihui Zhang, Yue Jiang, Yuhao Xu, Zenan Zhang, Yuanxing Zhou, Wiggin Zhou, Chayse Lian, Fengzong |
| contents | The generative capabilities of Large Language Models (LLMs) are rapidly expanding from static code to dynamic, interactive visual artifacts. This progress is bottlenecked by a critical evaluation gap: established benchmarks focus on algorithmic correctness and are blind to the visual fidelity and interactive integrity that define modern user experiences. To bridge this gap, we introduce ArtifactsBench, a new benchmark and paradigm for the automated, multimodal evaluation of visual code generation. Our framework programmatically renders each generated artifact and captures its dynamic behavior through temporal screenshots. This visual evidence, alongside the source code, is then assessed by a Multimodal LLM (MLLM)-as-Judge, which is rigorously guided by a fine-grained, per-task checklist to ensure holistic and reproducible scoring. We construct a new benchmark of 1,825 diverse tasks and evaluate over 30 leading LLMs. Our automated evaluation achieves a striking 94.4% ranking consistency with WebDev Arena, the gold-standard for human preference in web development, and over 90% pairwise agreement with human experts. This establishes ArtifactsBench as the first framework to reliably automate the assessment of human-perceived quality at scale. Our analysis provides a high-resolution map of the current SOTA, revealing that generalist models often outperform domain-specific ones. We open-source ArtifactsBench, including the benchmark, evaluation harness, and baseline results at https://artifactsbenchmark.github.io/, to provide the community with a scalable and accurate tool to accelerate the development of user-centric generative models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_04952 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ArtifactsBench: Bridging the Visual-Interactive Gap in LLM Code Generation Evaluation Zhang, Chenchen Li, Yuhang Xu, Can Liu, Jiaheng Liu, Ao Zhou, Changzhi Deng, Ken Wu, Dengpeng Huang, Guanhua Li, Kejiao Yi, Qi Xiong, Ruibin Hu, Shihui Zhang, Yue Jiang, Yuhao Xu, Zenan Zhang, Yuanxing Zhou, Wiggin Zhou, Chayse Lian, Fengzong Computation and Language Software Engineering The generative capabilities of Large Language Models (LLMs) are rapidly expanding from static code to dynamic, interactive visual artifacts. This progress is bottlenecked by a critical evaluation gap: established benchmarks focus on algorithmic correctness and are blind to the visual fidelity and interactive integrity that define modern user experiences. To bridge this gap, we introduce ArtifactsBench, a new benchmark and paradigm for the automated, multimodal evaluation of visual code generation. Our framework programmatically renders each generated artifact and captures its dynamic behavior through temporal screenshots. This visual evidence, alongside the source code, is then assessed by a Multimodal LLM (MLLM)-as-Judge, which is rigorously guided by a fine-grained, per-task checklist to ensure holistic and reproducible scoring. We construct a new benchmark of 1,825 diverse tasks and evaluate over 30 leading LLMs. Our automated evaluation achieves a striking 94.4% ranking consistency with WebDev Arena, the gold-standard for human preference in web development, and over 90% pairwise agreement with human experts. This establishes ArtifactsBench as the first framework to reliably automate the assessment of human-perceived quality at scale. Our analysis provides a high-resolution map of the current SOTA, revealing that generalist models often outperform domain-specific ones. We open-source ArtifactsBench, including the benchmark, evaluation harness, and baseline results at https://artifactsbenchmark.github.io/, to provide the community with a scalable and accurate tool to accelerate the development of user-centric generative models. |
| title | ArtifactsBench: Bridging the Visual-Interactive Gap in LLM Code Generation Evaluation |
| topic | Computation and Language Software Engineering |
| url | https://arxiv.org/abs/2507.04952 |