# What Does Post-Training Change in Multilingual Reasoning?

arXiv: https://arxiv.org/abs/2609.37104 · HTML: https://mlsanity.com/paper/2609.37104/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 19 of its 23 references that are in the map.
- First posted on arXiv 29 Sep 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 19 of 19)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [High-Dimensional Continuous Control Using Generalized Advantage Estimation](https://mlsanity.com/paper/1506.02438/) (2015)
- [Do Llamas Work in English? On the Latent Language of Multilingual Transformers](https://mlsanity.com/paper/2402.10588/) (2024)
- [Language Model Tokenizers Introduce Unfairness Between Languages](https://mlsanity.com/paper/2305.15425/) (2023)
- [Do Multilingual LLMs Think In English?](https://mlsanity.com/paper/2502.15603/) (2025)
- [PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts](https://mlsanity.com/paper/2504.18428/) (2025)
- [Wait, Wait, Wait... Why Do Reasoning Models Loop?](https://mlsanity.com/paper/2512.12895/) (2025)
- [Learn Globally, Speak Locally: Bridging the Gaps in Multilingual Reasoning](https://mlsanity.com/paper/2507.05418/) (2025)
- [Long Chain-of-Thought Reasoning Across Languages](https://mlsanity.com/paper/2508.14828/) (2025)
- [Think Natively: Unlocking Multilingual Reasoning with Consistency-Enhanced Reinforcement Learning](https://mlsanity.com/paper/2510.07300/) (2025)
- [When Models Reason in Your Language: Controlling Thinking Language Comes at the Cost of Accuracy](https://mlsanity.com/paper/2505.22888/) (2025)
- [The Token Tax: Systematic Bias in Multilingual Tokenization](https://mlsanity.com/paper/2509.05486/) (2025)
- [The Impact of Language Mixing on Bilingual LLM Reasoning](https://mlsanity.com/paper/2507.15849/) (2025)
- [ReasonXL: Shifting LLM Reasoning Language Without Sacrificing Performance](https://mlsanity.com/paper/2604.12378/) (2026)
- [CoCo-CoLa: Evaluating and Improving Language Adherence in Multilingual LLMs](https://mlsanity.com/paper/2502.12476/) (2025)
- [LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance](https://mlsanity.com/paper/2605.22567/) (2026)
- [Disentangling Language Roles in Multilingual LLM Task Execution](https://mlsanity.com/paper/2605.27649/) (2026)

## Abstract (arXiv)

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.