The second stage applies RL (Dr. GRPO) to recover reasoning quality lost during SFT while preserving target-language compliance, using a composite reward over verifiable math problems.

Model Details

Base model: open-thoughts/OpenThinker-7B
Target reasoning language: Italian
SFT dataset: toroe/ReasonXL-SFT
Training stage: SFT
Corresponding GRPO model: DGurgurov/OpenThinker3-7B-SFT-GRPO-IT

Full training details, evaluation results, and methodology will follow soon.

Citation

If you use this model, please cite:

bibtex

@misc{gurgurov2026reasonxlshiftingllmreasoning,
      title={ReasonXL: Shifting LLM Reasoning Language Without Sacrificing Performance},
      author={Daniil Gurgurov and Tom Röhr and Sebastian von Rohrscheidt and Josef van Genabith and Alexander Löser and Simon Ostermann},
      year={2026},
      eprint={2604.12378},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2604.12378},
}

DGurgurov/OpenThinker3-7B-SFT-IT

OpenThinker3-7B-SFT-IT

Stage 1 — SFT

Stage 2 — RL

Model Details

Citation

DGurgurov/
OpenThinker3-7B-SFT-IT