SwingArena: Competitive Programming Arena for Long-context GitHub Issue Solving

TL;DR

SwingArena is a novel evaluation framework for LLMs that simulates real-world software development workflows, including collaborative patch creation and validation through CI pipelines, using retrieval-augmented code generation for long-context handling.

Contribution

It introduces a realistic, interactive evaluation setup for LLMs in software development, incorporating retrieval-augmented code generation and CI validation, which advances beyond static benchmarks.

Findings

GPT-4o excels at patch generation

DeepSeek and Gemini focus on correctness in CI validation

Framework scales across multiple programming languages

Abstract

We present SwingArena, a competitive evaluation framework for Large Language Models (LLMs) that closely mirrors real-world software development workflows. Unlike traditional static benchmarks, SwingArena models the collaborative process of software iteration by pairing LLMs as submitters, who generate patches, and reviewers, who create test cases and verify the patches through continuous integration (CI) pipelines. To support these interactive evaluations, we introduce a retrieval-augmented code generation (RACG) module that efficiently handles long-context challenges by providing syntactically and semantically relevant code snippets from large codebases, supporting multiple programming languages (C++, Python, Rust, and Go). This enables the framework to scale across diverse tasks and contexts while respecting token limitations. Our experiments, using over 400 high-quality real-world GitHub issues selected from a pool of 2,300 issues, show that models like GPT-4o excel at aggressive patch generation, whereas DeepSeek and Gemini prioritize correctness in CI validation. SwingArena presents a scalable and extensible methodology for evaluating LLMs in realistic, CI-driven software development settings. More details are available on our project page: swing-bench.github.io