Benchmark RadarSWE Refactor Bench
SWE Refactor Bench is a benchmark for evaluating coding agents on whole-repository stack migrations. It comprises 20 migrations covering 4 types of technical debt, with a three-stage evaluation protocol measuring migration completeness and behavioral correctness: Migration Audit, Behavioral Tests, and Agentic Verification.
- Released
- 2026-08-24
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
The benchmark addresses the gap in evaluating code migration beyond mere test passing, preventing shortcut solutions. It provides a rigorous testbed for developing reliable coding agents for long-horizon refactoring tasks, with findings on agent capability across migration categories.
Motivation
Modern software systems accumulate technical debt over decades of development, which makes migration expensive and largely manual.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.