You can select the model you want to use from the model catalog in AIP. The ability to select a model !== the ability to use the model. Prompts are tightly fitted to the model, down to the specific model version. If you try to use PromptA designed for ModelA and use it with ModelB you will get a train wreck. This is a limitation baked into LLMs. The work around is a truly impressive engineering exercise that involves labeled data, evals, and frameworks like DSPy, model routers, etc. We use our own framework. And you still can't expect great results.