aiqbench analysis
This arXiv paper is squarely about LLM reasoning, specifically whether reinforcement post-training gains in English-centric models carry over to other languages. Its findings are practical rather than sensational: transfer varies widely by base model, target language, and training method, and a stronger English-centric starting model does not guarantee more efficient cross-lingual transfer. For readers building or evaluating multilingual models, the key takeaway is that a single parallel language in training gives the biggest boost, and that evaluation should report relative transfer efficiency alongside raw accuracy.
How Well Does Mathematical Reasoning Transfer Across Languages in English-Centric LLMs?
arXiv:2510.02272v2 Announce Type: replace Abstract: Recent advances in Reinforcement Post-Training (RPT) have substantially improved the mathematical reasoning capabilities of large language models (LLMs), but it remains unclear how well such gains transfer across languages. In this work, we study cross-lingual mathematical reasoning transfer in English-centric LLMs under controlled multilingual evaluation. We first evaluate a range of English-centric reasoning models on multilingual mathematical reasoning benchmarks and find that transfer is highly variable across initial model choice, target language, and training paradigm. Through controlled comparative experiments, we further find that stronger English-centric initialization does not necessarily lead to stronger relative transfer efficiency, even when it yields higher multilingual reasoning accuracy. Finally, we show that the largest improvement comes from moving beyond English-only supervision: adding a single parallel language yields a substantial gain, while further gains from additional languages are smaller but consistent. Overall, our results highlight the need to evaluate multilingual mathematical reasoning directly, using both absolute performance and relative transfer-efficiency metrics.
Title:How Well Does Mathematical Reasoning Transfer Across Languages in English-Centric LLMs?
Abstract:Recent advances in Reinforcement Post-Training (RPT) have substantially improved the mathematical reasoning capabilities of large language models (LLMs), but it remains unclear how well such gains transfer across languages. In this work, we study cross-lingual mathematical reasoning transfer in English-centric LLMs under controlled multilingual evaluation. We first evaluate a range of English-centric reasoning models on multilingual mathematical reasoning benchmarks and find that transfer is highly variable across initial model choice, target language, and training paradigm. Through controlled comparative experiments, we further find that stronger English-centric initialization does not necessarily lead to stronger relative transfer efficiency, even when it yields higher multilingual reasoning accuracy. Finally, we show that the largest improvement comes from moving beyond English-only supervision: adding a single parallel language yields a substantial gain, while further gains from additional languages are smaller but consistent. Overall, our results highlight the need to evaluate multilingual mathematical reasoning directly, using both absolute performance and relative transfer-efficiency metrics.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2510.02272 [cs.CL]
(or arXiv:2510.02272v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2510.02272 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 262 words