New arena shows coding agents can both create and solve tasks, with difficulty scaling with agent capability

Static coding benchmarks degrade over time as their tasks leak into training data and stronger agents eventually saturate them. The paper argues that benchmarks should instead aim to be self-scaling: tasks and evaluation criteria are created at evaluation time, so no content exists in advance to leak, and difficulty is set by the participating agents rather than fixed at release.

The authors introduce SWE-Duel, an arena where two coding agents evaluate each other over mirrored Red-Blue turns. In each turn, the Red agent works on a real repository across three context-isolated sessions. It proposes and implements a new feature with feature tests, embeds a subtle bug in that feature along with sealed bug-triggering tests, and submits a pull request that must pass the feature tests and the repository's existing tests while failing exactly the bug tests. A fresh session of the same agent must also be able to repair the bug blind, which rules out unsolvable tasks.

The Blue agent then reviews the pull request without being told a bug exists. It wins only if its revision passes all three test suites (feature tests, repository tests, and bug-triggering tests), and then the two agents swap roles.

The authors ran a tournament among 11 model-harness participants on 15 repositories in five languages for 1,500 turns. GPT-5.5 with Codex CLI ranked first as the strongest solver, repairing 98.3% of the defects it reviewed. Claude Opus 4.8 with mini-swe-agent ranked second as the strongest task author, with a 27.3% Red win rate.

Within one model family, tasks written by the newest version defeated a fixed-reference solver far more often than those of older versions (42.9% vs. 4.0%), providing preliminary evidence that task difficulty scales with task authoring capability.