# SWE-Duel benchmark pits coding agents against each other to resist training-data contamination

One agent ships a feature with a hidden bug and a rival agent must find and fix it, and only test execution decides the score, with no model judge.

- Published: 2026-10-12T06:16:38.065Z
- Canonical: https://polylog.news/ai/2026-10-12/swe-duel-benchmark-pits-coding-agents-against-each-other-to
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.CL](https://arxiv.org/abs/2610.12569), [PyPI: swe-duel](https://pypi.org/project/swe-duel/), [arXiv cs.AI](https://arxiv.org/abs/2610.12542)

A new paper on arXiv, "SWE-Duel: Towards a Self-Scaling Arena for Coding Agents", starts from a problem every evaluation team knows. Static coding benchmarks are fixed once released. Their tasks leak into training data over time, and strong…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-10-12/swe-duel-benchmark-pits-coding-agents-against-each-other-to (subscription information: https://polylog.news/pricing).