Switch Models Mid-Conversation with /model
Not every task needs the most powerful (and expensive) model. Switch on the fly with /model.
/model
This opens an interactive model picker. Or specify directly:
/model sonnet
The real pattern: think with Opus, execute with Sonnet. Use Opus for complex planning, architecture decisions, and hard debugging. Switch to Sonnet once you move into execution mode — applying changes, running tests, committing files. Opus is roughly 5x more expensive, and roughly 80% of a typical session is execution work that Sonnet handles equally well.
A practical session workflow:
# Start with Opus for the hard thinking
/model opus
"Design the database schema for a multi-tenant SaaS app"
# Switch to Sonnet for the implementation grunt work
/model sonnet
"Now create the migration files based on that schema"
You can also set the model at launch:
claude --model sonnet
A rough guide for when to use each:
- Opus — complex debugging, architecture decisions, multi-file refactors, subtle logic bugs
- Sonnet — boilerplate generation, simple edits, test writing, documentation, routine tasks
- Haiku — quick questions, formatting, simple lookups
The context window carries over when you switch, so Claude doesn't lose any conversation history.
Think with Opus. Execute with Sonnet. Same output quality, way more runway on your usage limits.
Log in to leave a comment.
When typing feels slow — describing a complex bug, explaining architecture, or thinking through a problem out loud — press Option+V to switch to voice input. Speak naturally and Claude Code transcribes your words into a prompt, so you can describe what you need at the speed of thought.
When Claude is heading down the wrong path — editing the wrong file, writing code you don't want, or giving a long explanation you don't need — press Escape to stop it immediately. You keep everything it did up to that point and can redirect with a new prompt.
Closed a session and realized you weren't done? Pass --continue (or -c) when launching Claude Code to pick up exactly where you left off — same context, same files, same conversation history — without re-explaining what you were working on.