_version_ 1866911324801335296
author Zhuo, Terry Yue
Jin, Xiaolong
Liu, Hange
Jiang, Juyong
Liu, Tianyang
Gong, Chen
Bishnoi, Bhupesh
Mishra, Vaisakhi
Suppa, Marek
Ziems, Noah
Utpala, Saiteja
Xu, Ming
Song, Guangyu
Li, Kaixin
Cao, Yuhan
Liu, Bo
Liu, Zheng
Abdurakhmanova, Sabina
Yu, Wenhao
Jia, Mengzhao
Yao, Jihan
Hamilton, Kenneth
Shridhar, Kumar
Vu, Minh Chien
Wang, Dingmin
Liu, Jiawei
Wang, Zijian
Liu, Qian
Hui, Binyuan
Risdal, Meg
Khaliq, Ahsen
Sood, Atin
Xing, Zhenchang
Ahmad, Wasi Uddin
Grundy, John
Lo, David
Zhu, Banghua
Du, Xiaoning
Scholak, Torsten
von Werra, Leandro
author_facet Zhuo, Terry Yue
Jin, Xiaolong
Liu, Hange
Jiang, Juyong
Liu, Tianyang
Gong, Chen
Bishnoi, Bhupesh
Mishra, Vaisakhi
Suppa, Marek
Ziems, Noah
Utpala, Saiteja
Xu, Ming
Song, Guangyu
Li, Kaixin
Cao, Yuhan
Liu, Bo
Liu, Zheng
Abdurakhmanova, Sabina
Yu, Wenhao
Jia, Mengzhao
Yao, Jihan
Hamilton, Kenneth
Shridhar, Kumar
Vu, Minh Chien
Wang, Dingmin
Liu, Jiawei
Wang, Zijian
Liu, Qian
Hui, Binyuan
Risdal, Meg
Khaliq, Ahsen
Sood, Atin
Xing, Zhenchang
Ahmad, Wasi Uddin
Grundy, John
Lo, David
Zhu, Banghua
Du, Xiaoning
Scholak, Torsten
von Werra, Leandro
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2510_08697
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle BigCodeArena: Unveiling More Reliable Human Preferences in Code Generation via Execution
Zhuo, Terry Yue
Jin, Xiaolong
Liu, Hange
Jiang, Juyong
Liu, Tianyang
Gong, Chen
Bishnoi, Bhupesh
Mishra, Vaisakhi
Suppa, Marek
Ziems, Noah
Utpala, Saiteja
Xu, Ming
Song, Guangyu
Li, Kaixin
Cao, Yuhan
Liu, Bo
Liu, Zheng
Abdurakhmanova, Sabina
Yu, Wenhao
Jia, Mengzhao
Yao, Jihan
Hamilton, Kenneth
Shridhar, Kumar
Vu, Minh Chien
Wang, Dingmin
Liu, Jiawei
Wang, Zijian
Liu, Qian
Hui, Binyuan
Risdal, Meg
Khaliq, Ahsen
Sood, Atin
Xing, Zhenchang
Ahmad, Wasi Uddin
Grundy, John
Lo, David
Zhu, Banghua
Du, Xiaoning
Scholak, Torsten
von Werra, Leandro
Software Engineering
Artificial Intelligence
Computation and Language
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.
title BigCodeArena: Unveiling More Reliable Human Preferences in Code Generation via Execution
topic Software Engineering
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2510.08697