Pick the right model and effort level in Claude Code

The picker only lists models your version knows, so an old install hides the newest ones.

1. 1New to Claude Code? On a Mac, Linux or WSL, run the command below. In Windows PowerShell, run irm https://claude.ai/install.ps1 | iex.
2. 2Already installed? Run claude update, or brew upgrade claude-code if you used Homebrew.
3. 3Open a new terminal window and run claude --version.
4. 4In your project folder, run claude and log in when asked.

curl -fsSL https://claude.ai/install.sh | bash

You should see a version number of 2.1.293 or higher, then a Claude Code session open in your project.

Opus 5.5 is the default on every Claude plan and on the API, and the picker is where you change it.

1. 1In your session, type /model and press Enter.
2. 2Pick a model. Press Enter to switch and save it as your default, or s to switch for this session only.
3. 3Run /status to check which model is active.

You should see the model you picked in /status, next to your account details.

Each model has its own cache, so the first message after a switch re-reads the whole chat. Switch at the start of a task.

A lighter model on routine work saves your limit and money for jobs that need more depth.

1. 1Not sure? Stay on opus. Anthropic suggests Opus 5.5 for most work, including long agent runs.
2. 2Daily coding with a clear scope: /model sonnet gives you Sonnet 5.5, faster and cheaper.
3. 3Simple tasks like a rename: /model haiku gives you Haiku 5.5, the fastest.
4. 4/model opusplan plans with Opus in plan mode, then writes the code with Sonnet.

/model opus
/model sonnet
/model haiku
/model opusplan

You should see a switch to the model you named, which also becomes your default.

API prices per million tokens in and out: Opus 5.5 is $4 and $20, Sonnet 5.5 is $2 and $10. Haiku 5.5 starts at $0.10 and $0.50, and Fable 5.1 is $10 and $50.

Effort sets how hard Claude thinks on each step, so it changes speed, cost and depth.

1. 1Type /effort to open the slider, or /effort high to set a level directly.
2. 2Use low for quick edits, medium for clear feature work and high when edge cases matter.
3. 3Save max for hard problems you hand off. Test it first, since it can overthink.
4. 4For one deeper answer, put ultrathink anywhere in that prompt.

/effort medium

You should see the level in the session header, for example "with medium effort".

Opus 5.5, Sonnet 5.5 and Haiku 5.5 start at medium. Fable 5.1 starts at high.

Fable 5.1 is built for long, hard, open problems, so give it a goal rather than a script.

1. 1Run /model fable, or start a session with claude --model fable.
2. 2If Claude Code asks you to approve usage credits, read the prompt and accept.
3. 3Pick a task harder than one you'd give other models, such as a bug nobody has pinned down.
4. 4Paste the prompt below, fill in the brackets and send it.

I'm working on [THE PROJECT] for [WHO IT'S FOR]. They need [WHAT THE RESULT LETS THEM DO].

Goal: [THE OUTCOME YOU WANT].

Scope the work first and ask me any questions you need. Then do it end to end. Before you report back, check each claim against a test or command you ran, and say what you couldn't verify.

You should see a few scoping questions, then a plan, then Fable working through the task on its own.

To be asked before a switch, run /config, select Switch models when a message is flagged and choose Ask each time.

A shared settings file starts the whole team on the same model and effort.

1. 1In your project, open or create .claude/settings.json.
2. 2Add the keys below, with your team's alias: opus, sonnet, haiku or opusplan.
3. 3Commit the file so teammates get it.
4. 4Start a new session and run /status. The Setting sources line should list the project file.

{
  "model": "opusplan",
  "effortLevel": "medium"
}

You should see new sessions in this project starting on the model and effort you set.

Settings files are strict JSON, so a comment or trailing comma breaks them.