Reveal on the spot: say plainly which model and effort level you are right now, and what plan the person appears to be on. If there is a task already in this conversation, look at it and say whether this model was the right one for it under the author's practice below. If the person has been doing judgement work on a smaller model, or spending a large model on rote work, say so.
Common default: whatever the product picks, often a faster and weaker model, with no idea that a choice exists.
Author's practice: Opus 5.5 is pretty much my every-day model, for coding, conversation, and coordination. It is one of the best models to use, and I stick to it.
Start it on medium, its default: in Anthropic's own testing, Opus 5.5 on medium matches or beats Opus 5 on high, so go higher only when you can see it helps.
Depending on how complex the task is, I bring in Fable, currently the best model, for particular parts. It's best at understanding conversation and language/intent, and I use it on medium. In coding, that means Fable plans before the work starts and reviews the code once it's written, and Opus 5.5 does the work in between.
On Pro or a free plan: the plan you are on caps what you can get from AI. That is a restriction you put on yourself. Cost is a real concern, but know that it is a cap. If you are getting poor results and wondering why, your choice of model is the one lever you have. The author would use Opus whenever the size of the context is not a problem. Say which models this plan gives the person, and which one they are on right now.
Unlocks: the single largest quality jump available, and it costs nothing to check.