Newest papers

RefCraft: Benchmarking Visual-Reference-Conditioned Agentic Generation through Web Development

Yuchi Wang, Rongyao Fang, Weihao Yu, Xilin Wei, Tiancheng Gu, Dingkang Yang, Hang Zhang, Dayiheng Liu, Hongsheng Li, Shuai Bai · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26826 · cited by 0 works (OpenAlex) · open-access PDF

Russian

Помимо базовых навыков использования инструментов и кодирования агентства, эффективное включение визуальных ссылок в генерирование агентов является не менее важным для применения агентов ИИ в реальных задачах производительности. Мы изучаем эту ключевую способность через разработку веб-страниц, практическую и легко оцениваемую среду, в которой существующие ориентиры остаются в значительной степени идеализированными: они обычно сосредоточены на реконструкции визуальных целей или генерировании веб-страниц из текстовых инструкций в одиночку, игнорируя более распространенный рабочий процесс разработки из визуальных примеров. В частности, мы вводим RefCraft, ориентир, состоящий из 300 реалистичных мультимодальных задач по разработке веб-страниц, которые сочетают инструкции на естественном языке с

Machine translation (Meta M2M100); the English original follows.

Abstract

Beyond basic tool use and coding skills, effectively incorporating visual references into agentic generation is equally essential for applying AI agents to real-world productivity tasks. We study this crucial capability through webpage development, a practical and readily evaluable setting in which existing benchmarks remain largely idealized: they typically focus on reconstructing visual targets or generating webpages from textual instructions alone, overlooking the more common workflow of developing from visual examples. Specifically, we introduce RefCraft, a benchmark comprising 300 realistic multimodal webpage-development tasks that combine natural-language instructions with screenshots from one to five reference webpages. Agents must selectively interpret and utilize these references while rendering, visually inspecting, and iteratively refining their outputs. We evaluate the resulting open-ended artifacts using independently verifiable keypoints and complementary programmatic and VLM judges. Across nine frontier (M)LLMs, the strongest system achieves an overall score of only 52.12 out of 100, revealing that visual-reference-conditioned agentic generation remains a substantial challenge for current models in practical webpage-development settings.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Benchmarking, Workflow, Focus (optics), Artificial intelligence, Web application

Is this paper real? This page was built from OpenAlex and arXiv records, not generated. Software can verify any reference in one call: GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.26826, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txt

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange