| name: SWE-Bench Pro | |
| description: SWE-Bench Pro is an advanced, 1,865-task benchmark for evaluating AI coding agents on realistic, complex, long-horizon software engineering tasks. It improves upon the original SWE-Bench by focusing on non-trivial patches across 41+ repositories, aiming to address data contamination, task and complexity. | |
| evaluation_framework: swe-bench-pro | |
| tasks: | |
| - id: SWE_Bench_Pro | |
| config: default | |
| split: test |