Yes definitely give this a go, for us with our use case davinci is prohibitively expensive in production so we literally cannot use it given the number of reque

Yes definitely give this a go, for us with our use case davinci is prohibitively expensive in production so we literally cannot use it given the number of requests we make - interestingly I saw some OpenAI documentation the other day at the YC event which basically said that with a large enough dataset (I recall >= 30k examples) all of the models (yes even ada) start to behave with similar performance, so bear that in mind!