I would prefer “Use Multiple Models” for “Ask” to see how each model would implement the changes. Then create 1 agent run based on the preferred response/combination of responses.
In most cases, there is nothing to choose from. You either take the best expensive model, or you save money by taking a good cheap one. Unless when an expensive model can’t handle it (gpt-5-high), you can try another expensive model (Sonnet 4.5).
