Hierarchical reasoning apearently beat o1 Model , being only 25M model? apearently Reasoning is much better then normal LLM architecture.