← Changelog
NewApps

Choose the assistant's model, and cap what it spends

The assistant always used Claude Sonnet 5.5, and the only limit on a run was its 60 model calls. Now the chip beside the composer picks its model for the app, Sonnet 5.5, Haiku 5.5 for quick small changes or Opus 5.5 for hard problems, and a team admin can cap what one run may spend and what the app's assistant may spend in a month. A run that reaches a cap pauses and says which one.

The model is set for the app and everyone on the team, by anyone with Edit on it, and takes over from the next reply. Continue after a run's cap gives it the same amount again; the month's cap waits for an admin or the next month.

More in Model and spend.