# Long Chain-of-Thought Reasoning Across Languages

arXiv: https://arxiv.org/abs/2508.14828 · HTML: https://mlsanity.com/paper/2508.14828/

## Key facts

- Cited by 19 papers in the map (data through 14 Oct 2026).
- Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning.
- Builds on 0 of its 53 references that are in the map.
- First posted on arXiv 20 Aug 2025.

## Placed in

- LLM reasoning / [Prompting and chain-of-thought reasoning](https://mlsanity.com/lr/prompting-and-chain-of-thought-reasoning/)

## Built on by (most cited first, 19 of 19)

- [Crosslingual On-Policy Self-Distillation for Multilingual Reasoning](https://mlsanity.com/paper/2605.09548/) (2026)
- [Unlocking Multilingual Reasoning Capability of LLMs and LVLMs through Representation Engineering](https://mlsanity.com/paper/2511.23231/) (2025)
- [Gained in Translation: Privileged Pairwise Judges Enhance Multilingual Reasoning](https://mlsanity.com/paper/2601.18722/) (2026)
- [Beg to Differ: Understanding Reasoning-Answer Misalignment Across Languages](https://mlsanity.com/paper/2512.22712/) (2025)
- [Rethinking the Multilingual Reasoning Gap with Layer Swap](https://mlsanity.com/paper/2605.26735/) (2026)
- [Less Languages, Less Tokens: An Efficient Unified Logic Cross-lingual Chain-of-Thought Reasoning Framework](https://mlsanity.com/paper/2604.20090/) (2026)
- [Benchmarking Large Language Models for Quebec Insurance: From Closed-Book to Retrieval-Augmented Generation](https://mlsanity.com/paper/2603.07825/) (2026)
- [Is continuous CoT better suited for multi-lingual reasoning?](https://mlsanity.com/paper/2603.08177/) (2026)
- [Answering the Wrong Question: Reasoning Trace Inversion for Abstention in LLMs](https://mlsanity.com/paper/2604.02230/) (2026)
- [HiMed: Incentivizing Hindi Reasoning in Medical LLMs](https://mlsanity.com/paper/2605.24635/) (2026)
- [Beyond Input Understanding: Diagnosing Multilingual Mathematical Reasoning with Directed Acyclic Trace Graphs](https://mlsanity.com/paper/2605.27715/) (2026)
- [AdaMame: A Training Recipe for Adaptive Multilingual Reasoning](https://mlsanity.com/paper/2606.15080/) (2026)
- [Native Multilingual Chain-of-Thought Reasoning in Low-Resource Southeast Asian Languages](https://mlsanity.com/paper/2608.00533/) (2026)
- [RP-OPSD: Reasoning-Pivot-Guided On-Policy Self-Distillation for Multilingual Reasoning Transfer](https://mlsanity.com/paper/2608.06347/) (2026)
- [Terminal-Bench-LILT: Multilingual Agentic Coding Benchmark Grounded in Language, Region, and Culture](https://mlsanity.com/paper/2608.28641/) (2026)
- [More Capable, Less Faithful: A Multilingual Analysis of Mathematical (Un)Solvability Detection in LLMs](https://mlsanity.com/paper/2608.30463/) (2026)
- [From Zero to Hero: An Open LLM Ecosystem for Armenian](https://mlsanity.com/paper/2609.03350/) (2026)
- [What Does Post-Training Change in Multilingual Reasoning?](https://mlsanity.com/paper/2609.37104/) (2026)
- [Multilingual GSM-Symbolic: What determines capability transfer across languages?](https://mlsanity.com/paper/2610.03367/) (2026)

## Abstract (arXiv)

While large reasoning models have shown remarkable ability to generate long chains-of-thought (CoTs) in English, we still lack understanding of how these long-form reasoning abilities transfer to the vast majority of the world's languages. In this work, we systematically investigate four key stages of model development--scaling, pretraining, post-training, and inference--to understand how long CoT capabilities extend beyond English. We compare two reasoning settings across nine non-English target languages: En-CoT, where models process target-language inputs, but reason in English; and Target-CoT, where models both process inputs and generate long CoTs in the target language. We find that scaling reasoning model size improves multilingual task performance in En-CoT, but Target-CoT performance lags behind. This gap widens for tasks requiring long, multi-step CoTs such as mathematical reasoning. Shifting to pretraining, we find that adding a specialized reasoning stage enhances En-CoT performance but degrades Target-CoT, whereas broad multilingual pretraining improves both modes simultaneously. Given the scarcity of high-quality reasoning traces in languages other than English, we explore synthetic data curation approaches for post-training. We demonstrate that fine-tuning on reasoning traces automatically translated from gold English traces outperforms fine-tuning on target-language traces distilled from large reasoning models. Finally, we report disparities in inference efficiency between languages and uncover language-specific failure modes in CoTs. We release models, datasets, and code to foster further research.