Open AI & Automation → Agents → Kai Support → Settings → Model. What you see there depends on your plan.
On the current Starter, Team and Pro plans, the row reads Picked by Gleap and there's nothing to choose. Kai Support is billed per resolution, or included up to your monthly included AI, so Gleap runs it on the model that answers best. The rest of this article applies to plans where you can choose.
Auto picks a model for the task and falls back to an alternative when one isn't available. That fallback is the practical argument for leaving it alone: a provider outage doesn't stop Kai from answering.
On older plans without model selection, the row reads Auto routing and can't be changed. An Unlock button next to it shows what's needed.
Where model selection is available, the row becomes a dropdown. Each entry carries its price per million tokens, input and output, so the list doubles as a cost comparison.
Below the dropdown you see a short description of the selected model and Cost per answer, an estimate based on about 12,000 input and 800 output tokens. Use that figure rather than the per-million rates when comparing; it's closer to what a real conversation costs.
Gleap maintains the list and updates it as models are released and retired. If a model you pinned later leaves the list, your choice keeps working and stays visible in the dropdown, so nothing breaks silently and you can move to a current model when you're ready.
How the answer reads: phrasing and how well it holds a longer thread together.
How it handles a question that needs several pieces of your documentation combined.
Cost per answer, which varies widely across the list.
Whether customers can send screenshots. Not every model reads images. If the one you picked can't, the Images switch under Images & voice is disabled and names it.
Start on Auto and change it only when you have a reason. The usual one is cost, once you can see what you're spending per answer. If you pin a model and your customers tend to send screenshots, check that it reads images.
If you do change it, read a handful of real conversations afterwards. Model behaviour on your own content is hard to predict from a description, and the difference usually shows in how the model handles questions your documentation only partly covers.
