We dropped State-0, a 8B chain-of-thoughts alternative of GPT-o1 which is trained on additional 40million CoT parameters to give better responses with core 'socratic instincts'.