SWE-Bench ProMax: Benchmarking Agents on Large-Scale Multilingual Code Refactoring

2026-08-10

AI Summary

Existing benchmarks for evaluating AI programming agents suffer from flawed test cases, low task complexity, and limited cross-file scope, rendering them inadequate for assessing agents’ capabilities in real-world, large-scale, multi-language code refactoring. This work proposes the first multi-language benchmark specifically designed for high-complexity, behavior-preserving refactoring tasks, comprising 170 expert-curated and manually rewritten commit instances from open-source projects across seven programming languages. On average, each task involves modifications spanning 11.4 files and 261.6 lines of code. High quality and real-world relevance are ensured through human-rewritten task descriptions and manually validated test cases. Experimental results show that even state-of-the-art models achieve a maximum pass rate of only 41.2% under two widely used agent frameworks, underscoring the benchmark’s challenging and unsaturated nature.

Abstract

As AI coding agents take on increasingly complex, long-horizon software engineering tasks, existing benchmarks are rapidly saturating and their evaluation quality has come under serious scrutiny: a recent audit found that nearly 60% of unsolved SWE-bench Verified instances contain flawed tests -- either overly narrow tests that reject correct solutions or overly broad tests that check unstated requirements -- and that frontier models can verbatim reproduce gold patches from training data. Code refactoring, which requires coordinated, behavior-preserving changes across many files, offers a substantially harder and more realistic test of agent capability, yet remains underserved by current benchmarks. We introduce SWE-Bench ProMax, an expert-curated, multilingual code refactoring benchmark of 170 instances drawn from real commits across seven programming languages (Python, Java, TypeScript, Go, C, C++, and Rust). Every instance undergoes rigorous, multi-stage curation that directly addresses the quality problems identified in prior benchmarks: issue descriptions are rewritten from scratch to provide precise, unambiguous specifications, and test suites are manually reviewed to remove overly narrow and overly broad tests. Tasks with insufficient complexity or limited cross-file scope are filtered out, yielding a benchmark of challenging, large-scale refactoring tasks that average 11.4 modified files and 261.6 lines of code per instance, substantially exceeding the scale of existing benchmarks. Experiments with frontier models under two agent scaffolds show that the best model achieves only 41.2% resolve rate, confirming that SWE-Bench ProMax presents a meaningful and unsaturated challenge for current AI coding agents. Our benchmark is available at https://huggingface.co/datasets/swe-bench-promax/SWE-Bench-ProMax.

Problem

• code refactoring
• AI coding agents
• benchmark quality
• multilingual software engineering
• large-scale code changes

Innovation

• code refactoring
• multilingual benchmark
• AI coding agents
• test suite curation
• large-scale software engineering