SWE-Duel: Towards a Self-Scaling Arena for Coding Agents
We introduce SWE-Duel, an arena in which two coding agents evaluate each other over mirrored Red-Blue turns.
ProofPaper ↗
Key points
- 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.
- 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.
Sources (2)
- [1]SWE-Duel: Towards a Self-Scaling Arena for Coding AgentsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 06:01 PM
We introduce SWE-Duel, an arena in which two coding agents evaluate each other over mirrored Red-Blue turns.
Static coding benchmarks are fixed at release.
- [2]SWE-Duel: Towards a Self-Scaling Arena for Coding AgentsHugging Face Daily Papers · Oct 8, 12:00 AM · same content
Extractive summary: sentences quoted from the sources.
Before this
- Oct 7, 2026[AINews] Claude Haiku 5.5 — better than GPT-6 Luna at the same pricing
- Oct 7, 2026AgentTime: Can Agents Estimate and Control Their Own Runtime?
- Oct 7, 2026WorldBench: Evaluating LLMs on Three.js Voxel World Generation
- Oct 7, 2026MIMESIS: Learning User Simulators as Training Environments for Interactive Agents
- Oct 6, 2026Show HN: OpenChart – OSS TradingView alternative with your own AI agent
- Oct 5, 2026Anthropic Subscriptions Offer 5x+ More Value Than OpenAI