Do findings generalize from open-weight LRMs to proprietary LRMs?

Ascertain whether empirical differences between reasoning in English versus the question’s language, established using open-weight Large Reasoning Models, generalize to closed/proprietary models such as OpenAI o1 and Gemini 2.5 Pro.

Background

All experiments in the paper are conducted with open-weight reasoning models (e.g., Qwen and DeepSeek variants). The results suggest stronger performance when reasoning in English and reveal translation-related errors when translating non-English inputs to English.

The authors explicitly state uncertainty about whether these findings hold for closed/proprietary models, naming o1 and Gemini 2.5 Pro, and identify this generalization as an unresolved issue.