I've tried it with various models. And 4o is really good given that it returns data at least 10 times faster. But if you ask it to fill out a Json document, o

I've tried it with various models. And 4o is really good given that it returns data at least 10 times faster. But if you ask it to fill out a Json document, o3 (or other reasoning models) is still better, more correct and predictable. Or at least, better enough to justify waiting a minute for the API call to return vs 3-5 seconds.