I’m setting up a production app that calls the OpenAI API and am unsure whether to target a fixed model snapshot or a moving model alias. Pinning sounds safer for predictable behavior, but it also means keeping track of upgrades myself.

For teams shipping changes regularly, how do you handle model version updates? Do you test a new snapshot in parallel before switching, or rely on the alias and monitor for changes? I’m especially interested in what has worked for apps where small shifts in tone or formatting matter to users.