New chats start on a default model, so you don't have to pick one every time. You set separate defaults for chats in Code projects and for other chats, and you can give the app's background helpers lighter, cheaper models. You can still change the model in any chat. See Choose a model and effort level.
Set the default for new chats
- Open Settings and go to Models, in the Agents group.
- Find the section you want:
- Default code model is for chats in worktrees, in Code projects.
- Default chat model is for new chats that aren't in a worktree, such as chats in Work in Unstoppable Code.
- Each section has a row for Claude, Codex, and, on macOS and Windows, Ollama. On each row, choose a model and an effort level. The line under the name shows whether you're signed in to that provider.
- Select the circle beside Claude, Codex, or Ollama to make it the overall default. New chats start on that row's model.
The AI Credits row sets the model to use when a chat runs on AI credits, such as when no subscription you've connected can run it. Leave it on Automatic to let the app pick, or choose a model. The line under it shows Available, No spendable credits, or Unable to check. To learn how credits work, see AI credits and usage limits.
Under each section's description, Last resolved shows which option the app used for the last new chat, such as Claude subscription or AI Credits.
If you've saved an Agent with Default model as its base model, it uses these defaults too. Each chat with that Agent takes the default when it starts. See Create and share custom agents.
Choose models for background helpers
The app uses a model for small jobs that run in the background. They rarely need your most capable model, so a faster or cheaper one can save usage. Under Feature models, choose a model and effort level for each:
- Chat title model names your chat tabs.
- Conversation handoff model writes the summary that carries a chat over when you switch between Claude and Codex, recover a session, or continue an archived chat.
- Cron generation model turns a schedule you describe in words, such as "every weekday at 9am", into a schedule the app can run.
- If Command generation model turns a plain description of a run condition into a command for scheduled workflows.
- Plan parsing and diagrams model reads tracked plans and draws their diagrams.
- Code review writes review annotations and diagrams.