For my application, I use several mini prompts that are cobbled together and then finally returned. I am noting that 4.5-preview is significantly slower insofar as Cloudflare’s 2 minutes time out before the completions are all done compared to gpt-4o-mini where my longest set of prompts are done in about 30 seconds.
When constructing prompts I always have to think about Cloudflare’s 2 minute window.
That said, the quality of the returned completions is definitely very good. I look forward to the 4.5-mini model that hopefully will move a bit faster. I understand there’s a lot more compute going on.