Intelligence Selector
An agent's intelligence level sets which models it prefers. Each level maps to an ordered list of models that administrators configure, so a higher level generally means stronger reasoning, usually with more latency. The selector appears in an agent's Settings and in the chat composer.
The platform defines five levels, from lightest to strongest: Lite, Standard, Max, Ultra, and Ultra Max. The names and descriptions shown in the selector can differ by deployment, and a deployment can offer fewer levels.
Choose a level
Start with a representative task and define what a correct result looks like. Compare the available levels on that task using result quality, required corrections, and completion time.
A stronger level may help with complex reasoning, unfamiliar workflows, or large amounts of context. It does not guarantee a correct result or a faster one. A lighter level may be enough for a narrow, repeatable workflow.
Change one level at a time, save when prompted, and inspect the agent's next work. A model level does not grant new tools, fix missing credentials, or bypass approvals. Check those blockers before raising the level.
Who controls the levels
- Members who can edit an agent's settings choose its level in Settings or the composer. The Meta Agent can change it on request.
- Administrators decide which models back each level. Assign a primary model and ordered alternatives per level in Organization AI providers.
- Operators set the default level for new agents and can restrict tools for a level. See Tool access policy.
If the selector is read-only or a level is locked, the selector shows the reason.
Review the result
After changing a level, review the agent's work in the chat timeline. Open Usage for consumption detail. See Usage.