SWE-Duel: Towards a Self-Scaling Arena for Coding Agents
Quick Answer
SWE-Duel introduces a self-scaling coding benchmark where agents evaluate each other in real-time, with GPT-5.5 achieving a 98.3% defect repair rate.
Quick Take
The framework allows dynamic task creation, ensuring no prior knowledge leaks, and shows that newer model versions create more challenging tasks, as evidenced by a 42.9% win rate against older versions.
Key Points
- SWE-Duel allows agents to create tasks at evaluation time, preventing knowledge leakage.
- GPT-5.5 with Codex CLI ranks first, repairing 98.3% of defects reviewed.
- Claude Opus 4.8 ranks second with a 27.3% win rate as a task author.
- Newer model versions create more difficult tasks, outperforming older versions significantly.
- The tournament involved 11 participants across 15 repositories in five programming languages.
DeepSignal Analysis
What happened
SWE-Duel presents a self-scaling coding benchmark where agents evaluate each other in real-time. The framework allows for dynamic task creation, preventing prior knowledge leaks. GPT-5.5 achieved a 98.3% defect repair rate, while newer models created more challenging tasks, evidenced by a 42.9% win rate against older versions.
Key evidence
- SWE-Duel enables coding agents to evaluate each other through mirrored turns, where tasks are generated at evaluation time.
- GPT-5.5, using Codex CLI, repaired 98.3% of defects it reviewed during the tournament.
- Tasks authored by newer model versions defeated a fixed-reference solver 42.9% of the time, compared to 4.0% for older versions.
Why it matters
The introduction of self-scaling benchmarks like SWE-Duel addresses the limitations of static benchmarks that become outdated as tasks leak into training data. This dynamic approach could lead to more robust evaluations of coding agents, ensuring they are tested against relevant and challenging tasks. The performance of newer models in creating difficult tasks suggests a potential for continuous improvement in AI capabilities.
What to watch
Paper Resources
📖 Reader Mode
~2 min readAbstract:Static coding benchmarks are fixed at release. Over time their tasks leak into training data, and stronger agents eventually saturate them. We argue that a benchmark should instead aim to be self-scaling: its tasks and oracles are created at evaluation time, so no content exists in advance to leak, and its difficulty is set by the participating agents rather than fixed at release. We introduce SWE-Duel, an arena in which 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, then embeds a subtle bug in that feature along with sealed bug-triggering tests. The resulting pull request is admitted only if (1) it passes the feature tests and the repository's existing tests while failing exactly the bug tests, and (2) a fresh session of the same agent can 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, and then the two agents swap roles. We run a tournament among 11 model-harness participants on 15 repositories in five languages for 1,500 turns. The leaders win in different ways. GPT-5.5 with Codex CLI ranks first as the strongest solver, repairing 98.3% of the defects it reviews. Claude Opus 4.8 with mini-swe-agent ranks second as the strongest task author, with a 27.3% Red win rate. Finally, within one model family, tasks written by the newest version defeat a fixed-reference solver far more often than those of older versions (42.9% vs. 4.0%), giving preliminary evidence that task difficulty scales with task authoring capability.
| Comments: | Under review as a conference paper at ICLR 2027 |
| Subjects: | Computation and Language (cs.CL); Software Engineering (cs.SE) |
| MSC classes: | I.2, D.2 |
| Cite as: | arXiv:2610.12569 [cs.CL] |
| (or arXiv:2610.12569v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12569 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Wei Minn [view email]
[v1]
Thu, 8 Oct 2026 18:01:00 UTC (755 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →The "10th Juror": Open-Set Standpoint Screening for Bureaucratic Bias Detection
MARS-Gov introduces a framework for detecting bureaucratic bias in Dutch government documents, achieving a new state-of-the-art F1 score of 0.880. This model outperforms existing zero-shot detectors by 20.2 points and reduces unnecessary interventions to just 2.5%. The framework's dynamic '10th juror' adapts to emerging biases, enhancing legal language processing.