metadata
library_name: transformers
license: mit
pipeline_tag: text-generation
Introduction to our ReasonFlux-Coders
We introduce ReasonFlux-Coders, trained with CURE, our algorithm for co-evolving an LLM's coding and unit test generation abilities.
- ReasonFlux-Coder-7B and ReasonFlux-Coder-14B outperform similarly sized Qwen Coders, DeepSeek Coders, and Seed-Coders, and naturally integrate into common test-time scaling and agentic coding pipelines.
- ReasonFlux-Coder-4B is our Long-CoT model, outperforming Qwen3-4B while achieving 64.8% efficiency in unit test generation. We have demonstrated its ability to serve as a reward model for training base models via reinforcement learning (see our paper).
Introduction
We propose CURE, a novel reinforcement learning framework that co-evolves LLM coder and unit tester to improve the overall coding ability of large language models. Trained on just 4.5 K samples, our ReasonFlux-Coder models outperform similarly sized Qwen Coder, DeepSeek Coder and Seed Coder. Also, this is the first open-source Coding-RL project to make everything publicly available—including models, evaluation benchmarks, training and testing datasets, and training codes!
Citation
@article{wang2025cure,
title={Co-Evolving LLM Coder and Unit Tester via Reinforcement Learning},
author={Wang, Yinjie and Yang, Ling and Tian, Ye and Shen, Ke and Wang, Mengdi},
journal={arXiv preprint arXiv:2506.03136},
year={2025}
}