The OctoThinker family is built on carefully studied mid-training insights, starting from the Llama-3 family, to create a reinforcement learning–friendly base language model.
OctoThinker-1B-Short-Zero is trained using the R1-Zero-style reinforcement learning technique, starting from OctoThinker-1B-Short-Base without any supervised fine-tuning (SFT).
Training Recipe for OctoThinker-1B-Short-Base
Data Pipeline
Evaluation Results of OctoThinker-1B-Base Series
Note that we adopt the few-shot prompting evaluation for these base language models.
Data Pipeline
RL Training Dynamics of OctoThinker-1B-Zero Series
Data Pipeline
More about OctoThinker
Data Pipeline
Citation
Check out our paper for more details. If you use our models, and datasets or find our work useful, please cite
@article{wang2025octothinker,
title={OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling},
author={Wang, Zengzhi and Zhou, Fan and Li, Xuefeng and Liu, Pengfei},
year={2025},
journal={arXiv preprint arXiv:2506.20512},
note={Preprint}
}
