Leave it on AUTO.
That is not a cop-out, and it is not us hiding the choice. AUTO is cheaper and usually better than picking one yourself, for a reason worth understanding.
What AUTO actually does
It routes each step of the work to a different model inside the same answer.
A single reply might use a fast cheap model to do the searching and reading, then switch to Claude to do the writing. You pay the expensive rate only for the part that needed the expensive model.
Pick Claude Opus by hand and you pay the top rate for the searching too, for no gain, because the searching was never the part that needed a frontier model.
What is in the picker
Four families: Claude, GPT, Gemini and Grok, each with a cheaper fast option and a stronger slower one. The picker shows the current version of each with a one-line description, plus its speed, its relative intelligence and its credit cost.
The list changes as new models ship, which is why it is worth reading in the app rather than memorising from here.
When overriding is worth it
You want the strongest possible writing on something long. The top Claude model is the best writer available and the difference shows on a long script or a careful analysis. It costs the most credits.
You want a fast rough draft and do not care about polish. The cheap fast models are a fraction of the cost and fine for short edits, rewording, or a quick list.
You want current-events framing. Grok is strongest on real-time knowledge, so it is the one to reach for when the angle depends on something happening now.
You are doing heavy research and want depth over speed. The bigger Gemini model handles research-heavy work at a mid cost.
The spread is large
The gap between the cheapest and most expensive model is very wide, per word written. That is the whole reason AUTO exists, and the reason a month of scripts on the top model costs many times what the same work costs on AUTO.
If you want to know what any of this costs you specifically, the usage row under each answer shows the credits it used.
What the model does not change
Your voice, your brand and your memories apply whichever model runs. Those come from My Brand, not from the model.