Small coding agent learns efficiently using self-made programming tasks
FrogNano: Training a 4B Coding Agent via Online Task Synthesis
Artificial Intelligence
Summary
Creating helpful computer programs often needs powerful systems, but this work shows a smaller program can learn effectively by practicing on tasks it creates itself. The authors built FrogNano, a compact coding agent trained only with synthetic programming challenges generated online based on what it can currently handle. This approach helped FrogNano improve without relying on larger models for guidance. The research highlights that making tasks suited to the learner's current ability is important for success, and suggests smaller, efficient coding agents can work well even with limited hardware.
coding agentsoftware engineering tasksreinforcement learningsynthetic taskstask synthesislearnability frontierpost-trainingresource-constrained environments
Authors
Minseon Kim, Zhengyan Shi, Emiliano Penaloza, Christopher Cui, Roger Creus Castanyer, Maryam Hashemzadeh, Isadora White, Jonathan Light, Jeonghye Kim, Matheus Pereira, Darya Moldavskaya, Chinmay Singh, Fabio Vera, Baolin Peng, Xingdi Yuan, Marc-Alexandre Côté, Alessandro Sordoni
Abstract
We present FrogNano, a 4B coding agent designed to tackle software engineering (SWE) tasks efficiently and effectively, even under resource-constrained environments. It is post-trained exclusively via RL on around 1,500 SWE environments with synthetic tasks. A key ingredient for improving performance is an online task synthesis pipeline that creates tasks calibrated to the frontier of learnability for the current checkpoint. This report provides evidence that competitive small coding agents can be trained with synthetic tasks alone, without traditional distillation from larger models, and that generating tasks at the learnability frontier of the current agent is important. We report details on the training methodology, evaluations across diverse environments, and in-depth analyses, serving as a foundation for our ongoing exploration of lightweight yet capable coding agents that can run on minimal hardware.