CoSPlay: Cooperative Self-Play at Test-Time with Self-Generated Code and Unit Test

Zhangyi Hu, Chenhui Liu, Tian Huang, Jindong Li, Yang Yang, Jiemin Wu, Zining Zhong, Menglin Yang, Yutao Yue · 2026

Read CoSPlay: Cooperative Self-Play at Test-Time with Self-Generated Code and Unit Test by Zhangyi Hu, Chenhui Liu, Tian Huang, Jindong Li, Yang Yang, Jiemin Wu, Zining Zhong, Menglin Yang, Yutao Yue (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23491.txt.

Introduction

Hu and colleagues present CoSPlay, a training-free test-time method that improves code and unit tests without ground-truth tests. Reward training needs costly ground-truth tests, while test-time methods using self-made tests suffer when those tests are noisy or tied to bad code. CoSPlay explores varied ideas, spots failure modes, and uses pass-count signals from a code-test matrix to prune or fix weak codes and refresh tests. On four benchmarks it lifts Qwen2.5-7B-Instruct accuracy from 22.1 to 33.2 percent.

Catalog record

CREATORZhangyi Hu, Chenhui Liu, Tian Huang, Jindong Li, Yang Yang, Jiemin Wu, Zining Zhong, Menglin Yang, Yutao Yue
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE146104
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.23491

Browse the full library · About the Tower of Babel