AAUGS/UI

[ICLR 2026] Code Aesthetics with Agentic Reward Feedback

<div align="center"> <a href='https://bangx7.github.io/code-aesthetics/'><img src='https://img.shields.io/badge/Project-Page-Green'></a> <a href="https://huggingface.co/SamuelBang/AesCoder-4B"><img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-ffc107?color=ffc107&logoColor=white"/></a> <br> <a href="https://arxiv.org/abs/2510.23272"><b>Paper Link</b>👁️</a> </div> <div align="center"> <p> ^1,2<a href="https://bangx7.github.io" target="blank">Bang Xiao</a>^#,</span> <span class="author-block"> ^1,3<a href="https://github.com/JackLingjie" target="blank">Lingjie Jiang</a>^#,</span> <span class="author-block"> ¹<a href="https://www.microsoft.com/en-us/research/people/shaohanh/" target="blank">Shaohan Huang</a>^✉,</span> <span class="author-block"> ¹<a href="https://www.microsoft.com/en-us/research/people/tengchaolv/" target="blank">Tengchao Lv</a>, </span> <span class="author-block"> ¹<a href="https://www.microsoft.com/en-us/research/people/yupanhuang/" target="blank">Yupan Huang</a>, </span> <span class="author-block"> ¹<a href="https://yushuiwx.github.io/" target="blank">Xun Wu</a>, </span> <span class="author-block"> ¹<a href="https://www.microsoft.com/en-us/research/people/lecu/" target="blank">Lei Cui</a>, </span> <span class="author-block"> ¹<a href="https://www.microsoft.com/en-us/research/people/fuwei/" target="blank">Furu Wei</a> </span> </p> <p> ¹Microsoft Research Asia &nbsp;&nbsp; ²Zhiyuan College, Shanghai Jiao Tong University &nbsp;&nbsp; ³Peking University<br> ^#Equal Contribution ^✉Corresponding author </p> </div>

For the codebase, refer to: https://github.com/bangx7/code_aesthetics

🎉 News

• —_[2026.01.12]_: Release the AesCode dataset.
• —_[2025.10.29]_: Release the AesCoder-4B model.
• —_[2025.10.27]_: Release the Project Page and the Arxiv version.

📷 Abstract

Large Language Models (LLMs) have become valuable assistants for developers in code-related tasks. While LLMs excel at traditional programming tasks such as code generation and bug fixing, they struggle with visually-oriented coding tasks, often producing suboptimal aesthetics. In this paper, we introduce a new pipeline to enhance the aesthetic quality of LLM-generated code. We first construct AesCode-358K, a large-scale instruction-tuning dataset focused on code aesthetics. Next, we propose agentic reward feedback, a multi-agent system that evaluates executability, static aesthetics, and interactive aesthetics. Building on this, we develop GRPO-AR, which integrates these signals into the GRPO algorithm for joint optimization of functionality and code aesthetics. Finally, we develop OpenDesign, a benchmark for assessing code aesthetics. Experimental results show that combining supervised fine-tuning on AesCode-358K with reinforcement learning using agentic reward feedback significantly improves performance on OpenDesign and also enhances results on existing benchmarks such as PandasPlotBench. Notably, our AesCoder-4B surpasses GPT-4o and GPT-4.1, and achieves performance comparable to large open-source models with 480B-685B parameters, underscoring the effectiveness of our approach.

To-do List

• —[x] Release paper and project page
• —[x] Release our AesCoder model
• —[x] Release AesCode dataset
• —[x] Release code

Note: This is the version of Qwen3-Coder-480B-A35B-Instruct-FP8 generated AesCode dataset for only webpage design. For GPT-5 generated data, please stay tuned!

Dataset Overview

This dataset contains 5 categories of webpage design:

• —General Website: ~100k
• —Data Visualization: ~30k
• —3D Design: ~30k
• —Game Development: ~30k
• —UI Component: ~30k

&#x1F4DA; Citation

If you find this codebase useful for your research, please use the following entry.

@misc{xiao2025codeaestheticsagenticreward,
      title={Code Aesthetics with Agentic Reward Feedback}, 
      author={Bang Xiao and Lingjie Jiang and Shaohan Huang and Tengchao Lv and Yupan Huang and Xun Wu and Lei Cui and Furu Wei},
      year={2025},
      eprint={2510.23272},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.23272}, 
}