selected samples. Notably, the Marco-o1 model has achieved considerable breakthroughs in multilingual applications. Testing has shown improvements in accuracy, with increases of 6.17% in the English MGSM dataset and 5.60% in its Chinese equivalent. The model's performance in translation tasks is especially remarkable, adeptly managing colloquial expressions and understanding cultural contexts. A standout innovative feature of Marco-o1 is its application of varied action granularities within the MCTS framework. This empowers the model to explore reasoning pathways with differing levels of detail, ranging from generalized st