CodeMidas: Scaling Agentic Coding RL Environments from Code Itself
CodeMidas turns source code into 5,545 RL tasks and improves MiMo-V2.5 on five coding benchmarks.
CodeMidas is an agentic pipeline that turns implemented functionality in open-source code into executable reinforcement-learning tasks using only source code. The dataset has 5,545 training tasks from 3,185 codebases spanning 23 languages and 15 domains. Training MiMo-V2.5 with GRPO improved all five reported benchmarks, including DeepSWE by 11.7%, ProgramBench by 17%, and Terminal-Bench v2.1 by 8.5%. Ablations show more high-quality tasks and stronger exploration and self-verification.
- 5,545 tasks drawn from 3,185 codebases and 23 languages
- Pipeline uses source code only, not issues or commits
- MiMo-V2.5 trained with GRPO improved all five benchmarks
- Gains include DeepSWE 11.7%, ProgramBench 17%, Terminal-Bench 8.5%
Full article214 words · extracted from arxiv.org · click to collapse
Training capable coding agents via reinforcement learning (RL) requires diverse tasks with reliable verifiers. Open-source codebases offer a rich source of such tasks, while existing methods typically rely on development artifacts such as issues and commits, limiting the range of tasks that can be extracted. To better scale RL environments, we present CodeMidas, an agentic pipeline that turns implemented functionality in existing codebases into executable RL environments using source code as its only task-specific input. CodeMidas allocates agentic compute to every stage of environment construction: agents explore implemented functionality to formulate behavioral specifications, construct tests grounded in execution of the original code, and validate and filter candidate tasks through execution checks and repeated solution rollouts. The resulting dataset has 5,545 training tasks from 3,185 open-source codebases spanning 23 programming languages and 15 technical domains. Training MiMo-V2.5 on these tasks with GRPO improves performance on all five diverse benchmarks, covering issue repair (DeepSWE + 11.7%), whole-program construction (ProgramBench +17%), and terminal work (Terminal-Bench v2.1 +8.5%). Ablations show that increasing the number of high-quality training tasks improves performance. Trajectory analysis shows the RL-trained agent demonstrates better behaviors like increasing codebase exploration and more diverse self-verification. These results establish source code as a scalable foundation for constructing RL environments that improve coding agents across diverse software tasks.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.22068