BigCodeArena: Unveiling More Reliable Human Preferences in Code Generation via Execution
Abstract
Crowdsourced model evaluation platforms, such as Chatbot Arena, enable real-time evaluation from human perspectives to assess the quality of model responses. In the coding domain, manually examining the quality of LLM-generated content is extremely challenging, as it requires understanding long chunks of raw code and deliberately simulating code execution. To this end, we introduce BigCodeArena, an open human evaluation platform for code generation backed by a comprehensive and on-the-fly execution environment. Built on top of Chatbot Arena, BigCodeArena enables the execution of LLM-generated code and allows humans to interact with the execution process and outcomes. We collected over 14,000 raw code-centric conversation sessions across 10 widely used LLMs, spanning 10 languages and 8 types of execution environments. Among these conversations, we identified more than 4,700 multi-turn samples with pairwise human preferences. Further analysis uncovers underexplored preferences of LLMs in fine-grained domains characterized by tasks, languages, and frameworks. To systematically examine code understanding and generation capabilities of frontier LLMs, we curated two benchmarks based on the collected data, namely BigCodeReward and AutoCodeArena. For BigCodeReward, we post-processed the 4,700 conversations and evaluated the consistency between reward models and human preferences. The evaluation shows that most LLMs have superior performance in judging coding preferences when the execution results are available. Inspired by these findings, we propose AutoCodeArena, an automatic Elo rating benchmark designed to assess the coding quality of LLMs without human involvement. We find that proprietary LLMs like GPT-5, Claude-Sonnet-4, and Claude-Opus-4 still lead in code generation performance among recent emerging models.
Cite
@article{arxiv.2510.08697,
title = {BigCodeArena: Unveiling More Reliable Human Preferences in Code Generation via Execution},
author = {Terry Yue Zhuo and Xiaolong Jin and Hange Liu and Juyong Jiang and Tianyang Liu and Chen Gong and Bhupesh Bishnoi and Vaisakhi Mishra and Marek Suppa and Noah Ziems and Saiteja Utpala and Ming Xu and Guangyu Song and Kaixin Li and Yuhan Cao and Bo Liu and Zheng Liu and Sabina Abdurakhmanova and Wenhao Yu and Mengzhao Jia and Jihan Yao and Kenneth Hamilton and Kumar Shridhar and Minh Chien Vu and Dingmin Wang and Jiawei Liu and Zijian Wang and Qian Liu and Binyuan Hui and Meg Risdal and Ahsen Khaliq and Atin Sood and Zhenchang Xing and Wasi Uddin Ahmad and John Grundy and David Lo and Banghua Zhu and Xiaoning Du and Torsten Scholak and Leandro von Werra},
journal= {arXiv preprint arXiv:2510.08697},
year = {2025}
}
Comments
Built with love by the BigCode community :)