SWE-Bench Pro V2: A Cleaner, Harder-to-Game Coding Benchmark
Soham Dan, Jeff Da, Sanyam Bhutani (Reflection AI) and the Scale AI team release SWE-Bench Pro V2, a revised public split in which agentic audits, expert adjudication and blind re-implementation repaired or retired 480 of the original 731 tasks.
Ask this paper
Human audit cost. 23 contracted engineers spent 404 hours, each in a language they write professionally. Every revised instruction was checked by a second engineer solving the task from the text alone.
Harder to game. Network access is restricted to the model endpoint and each patch is re-graded in a clean container, so agents cannot fetch upstream fixes or get credit for changes outside their patch.
Results. On the 642 V2 tasks frontier agents resolve up to 99.4%, while other systems stay separated. A 51-task Hard split spreads systems from 52.9% to 98.0%.
Maintenance lessons. As agents improve they reach task issues earlier agents never touched, so benchmarks need ongoing audits rather than a one-time release.
Abstract
SWE-Bench Pro is a widely used benchmark of long-horizon software engineering tasks for coding agents. As agents improve, they reach task issues earlier agents never touched and interact with the evaluation environment in ways the original harness did not anticipate. We present SWE-Bench Pro V2, a revised public split built to keep the benchmark accurate as capabilities advance. Agentic audits, expert adjudication and blind re-implementation repaired or retired 480 of the 731 tasks: 23 contracted engineers spent 404 hours, each working in a language they write professionally, and every revised instruction was verified by a second engineer solving the task from the text alone. A hardened protocol restricts network access to the model endpoint and re-grades every patch in a clean container, so agents cannot retrieve upstream fixes or get credit for changes outside their patch. On the 642 V2 tasks frontier agents resolve up to 99.4% while other systems stay clearly separated, and a 51-task Hard split spreads systems from 52.9% to 98.0%. We release the benchmark and distill lessons for maintaining agent benchmarks over time.