I use o3-mini-high in Aider, where I want a model to employ reasoning but not put up with the latency of the non-mini o1.