ReasonXL: Shifting LLM Reasoning Language Without Sacrificing Performance

TL;DR

This paper introduces ReasonXL, a large-scale multilingual reasoning dataset, and demonstrates how LLMs can be adapted to reason in different languages with minimal performance loss.

Contribution

It provides the first parallel corpus for cross-lingual reasoning, a two-stage adaptation pipeline, and an analysis of how reasoning representations shift across model layers.

Findings

Models can be effectively adapted to new languages with minimal knowledge loss.

RLVR achieves greater behavioral divergence with smaller parameter updates than SFT.

Early layers encode language identity, while upper layers adapt reasoning capabilities.

Abstract

Despite advances in multilingual capabilities, most large language models (LLMs) remain English-centric in their training and, crucially, in their production of reasoning traces. Even when tasked with non-English problems, these models predominantly reason in English, creating a fundamental mismatch for non-English usage scenarios. We address this disparity directly with three contributions. (i) We introduce ReasonXL, the first large-scale parallel corpus of cross-domain reasoning traces spanning five European languages (English, German, French, Italian, and Spanish), with over two million aligned samples per language, each comprising prompts, reasoning traces, and final outputs, enabling direct supervision of language-specific reasoning. (ii) Using ReasonXL, we demonstrate that LLMs can be adapted to reason entirely in a desired target language, using a simple two-stage pipeline of supervised fine-tuning (SFT) followed by reinforcement learning with verifiable rewards (RLVR). The resulting models match or exceed baseline performance, with minimal loss in general knowledge and broadly preserved cross-lingual transfer. (iii) We conduct an extensive representational analysis of the adaptation and find a clear functional division across model depth: early layers contain an activation bottleneck that causally determines language identity, while upper layers concentrate the weight and activation changes driven by adaptation. We further find that RLVR achieves greater behavioral divergence from the base model with smaller parameter updates than SFT, suggesting a more efficient representational rerouting despite much smaller weight updates.