@OpenAI good point, but sometimes I just abort the thinking process because its writing is significantly worse than the fast model. Same as o3 vs 4o.
And very often, I need to rewrite a yet well-thought answer.
When will thinking models be at least as good as fast ones regarding writing?