My team measured Davinci and Curie response time during the day to understand the occasional long response times we get in our application (tested with 1 token prompt):
You can indeed get a dedicated instance, they start to make commercial sense if you are using ~450M tokens per day, you can reach out to get more information here
