17 Feb 2025

o1 and o3 mini join the existing "reasoning" models available on AnyModel: o1 preview, o1 mini and DeepSeek R1. But what makes these models different to other LLMs? 

"Reasoning" models spend time "thinking" before producing results. This process, also known as "test time compute", involves the models generating a "chain of thought" before answering. These chains allow the model to break the users request down into smaller sub-tasks which the model can carry out in order. 

This innovation means that AI models can essentially prompt themselves, enabling them to carry out more complex tasks and perform better results with simpler prompting.