These stages establish a constructive post-training path from English-pivoted capability to multilingual reasoning that is reliably delivered, and jointly track correctness, language adherence, termination, and delivery efficiency.
Abstract
Open-source reasoning models provide unequal access to reasoning capability across languages. When a model can solve a problem but cannot deliver a complete solution in the user's language, language becomes an access barrier rather than merely a source of performance variation. We audit Qwen3 checkpoints on competition-mathematics tasks in eleven languages. Across the ten non-English languages, only 15.4-17.9% of problems receive a correct, terminating solution with visible reasoning in the requested language in any of 16 samples, compared with 92.9% in English. To identify the source of this disparity, we evaluate thirteen endpoints from one model family, spanning released checkpoints, multilingual supervised fine-tuning (SFT) at two scales, controlled SFT ablations, and three reinforcement-learning (RL) reward formulations. We jointly track correctness, language adherence, termination, and delivery efficiency. The dominant bottleneck shifts across post-training stages. Released models often reason in English. Multilingual SFT restores target-language reasoning, but accuracy declines across multilingual, English-only, and single-language SFT runs, showing that this cost is not specific to multilingual mixing; non-English reasoning traces additionally become prone to non-terminating loops. RL restores termination in both arms at no cost in accuracy, but only the arm whose reward includes a language term delivers: rewarding correctness alone returns the model to English. Together, these stages establish a constructive post-training path from English-pivoted capability to multilingual reasoning that is reliably delivered.
It is shown the path to generalizing L2 reasoning to held-out languages goes through broader language coverage, readily available multilingual non-reasoning data, and a sufficient English reasoning backbone, indicating that reasoning is a language-agnostic behavior that can be transferred across typologically diverse l...
Mehrnaz Mofakhami, Ananya Sahu, Alejandro R. Salamanca et al.· 0 citations
This work quantifies cross-lingual skill inconsistency orthogonally from knowledge and general benchmark performance via multilingual self-play, and shows that skill discrepancies are a measurable major roadblock in the development of truly multilingual models.
Bobby Cheng, Adam Gaber, Zhengzhe Liu et al.· 1 citation
It is found that, in this benchmark, supervised fine-tuning (SFT) provides a strong baseline, substantially improving argument language consistency and end-to-end function call accuracy and, under consistent model selection, SFT achieves performance comparable to, and sometimes exceeding more complex reinforcement lear...
Siddharth Chauhan, Thomas Butler, Abhishek Singhania et al.· 0 citations
Reliability-Aware Adaptive Inference (RAAI), a training-free test-time framework that performs Expected Calibration Error-aware routing and prompt fusion, and uses a mid-layer Risk Index to gate sequential reasoning, allocating compute only when it is likely to help and suppressing harmful bilingual redundancy, is addr...
Ekata Mitra, Ameeta Agrawal· Conference on Empirical Meth...· 0 citations
A unified reasoning reward model supporting multiple languages and evaluation paradigms, UniRRM uses a staged reasoning chain to dynamically generate task-generic and instruction-specific criteria, enabling fine-grained, input-adaptive judgments while maintaining consistency across languages.
Peng Lai, Yi-Chao Du, Junchao Wu et al.· 1 citation
It is found that training to reason in the native language often leaves only a small gap to training for English reasoning, and RLVR beyond English can provide broad crosslingual gains, but also requires broad evaluation to detect language-specific regressions.
Konstantin Dobler, Federico Scozzafava, Jonathan Janke et al.· 0 citations