Skip to main content

Choosing a Model

The model picker is the rightmost control in the composer footer. It applies to the message you are about to send, not to the whole conversation — so you can change model between turns in the same chat.

Auto

Auto is offered first in the list and is the default. Its hint:

Model chosen per task — the cheapest that can do the job

The order matters. Auto filters by capability first, and only then by cost. A saving never routes a message to a model that would fail it:

Your messageWhat Auto must pick
Contains an imageA vision-capable model
Is sent in Build modeA tool-capable model

Because the choice is made per task, a cheap question and an expensive build in the same conversation can land on different models. That is Auto working, not drifting.

Cost

What you seeMeaning
MixedAuto — the cost depends on which model each message needs
Low / Medium / HighA specific model's band

Hover a specific model and the tooltip shows the exact multiplier. The bands come from the relative cost ratings an administrator sets in Settings → AI → Setup, so what counts as Low on your instance reflects your own model line-up. See Model Providers.

Reading the list

Models are grouped under a vendor badge. Two icons mark capabilities:

IconMeaning
EyeThe model can read images
GlobeThe model can search the web

An unavailable model is shown greyed out with the reason rather than hidden. You can see it exists and why you cannot use it — a disabled model looks different from a model that was never configured, and the reason tells you what to ask an administrator for.

Your choice is honoured

Picking a specific model is honoured exactly. Nothing silently substitutes another one.

If the model you named is unavailable, the run reports that rather than answering on something else. An answer you get is always from the model you asked for.

Auto can be missing

Auto is unavailable if an administrator has taken every model out of the Auto pool. Pick a specific model instead.

When to override Auto

Auto is the right answer most of the time. These are the cases worth thinking about:

SituationWhat to do
Pasting a screenshotPick a model with the eye icon, or leave Auto — it will pick a vision-capable model for you
A long, complex buildPick a stronger model. Auto optimises for the cheapest model that can do the job, which is not always the one that does it best
A cheap, repetitive questionPick a Low cost model and stay on it for that stretch of the conversation
You need current information from the webPick a model with the globe icon
Web search is enabled per model

The globe appears only on models an administrator has turned web search on for. No model reaches the internet otherwise. See Model Providers.

Verify it worked

  • The picker shows Auto with cost Mixed on a fresh message.
  • Hovering a specific model shows its exact cost multiplier.
  • Changing the model on one message and sending another leaves the first answer untouched — the setting is per message.
  • Send an image on a model with the eye icon and the reply describes what is in it.

If it didn't work

SymptomCauseFix
No models at all, or "No AI model is available"No provider is configured, or every model is disabledAn administrator needs to add a provider and enable at least one model — see Model Providers
The model I want is greyed outThe reason is shown next to it — usually disabled, not licensed for this instance, or missing a working provider credentialRead the reason, then ask an administrator for that specific change
Auto is not in the listAn administrator has taken every model out of the Auto poolPick a specific model. Ask an administrator to return models to the pool if you want Auto back
A model cannot read my screenshotThe model has no eye icon — it is not vision-capableSwitch to a model with the eye icon, or use Auto, which picks one when the message has an image
The reply says the model I picked is unavailableIt became unavailable between your picking it and sendingPick another model, or Auto. Nothing was answered on a substitute
No globe on any modelWeb search is off for every modelAsk an administrator to enable it on a specific model — see Model Providers
Costs look wrong for the models we runThe relative cost ratings need adjustingAn administrator sets them in Settings → AI → Setup — see Policy, Budgets and Limits

Next