Show Thinking in Claude Code: Verbose & Extended Mode

Ctrl+O toggles verbose output in Claude Code. When active, you see everything the model processes before and while it responds: tool call details, execution traces, and — crucially — the extended thinking blocks that reveal the model's internal reasoning.

This shortcut flies under the radar, but it's one of the most valuable when you're working with a thinking model. Being able to read the reasoning in real time turns you into a live reviewer — you can catch hallucinations, wrong assumptions, or flawed logic before they materialize into code changes.

This matters most when Claude is working on something complex. The thinking trace shows you how the model is approaching the problem. If you spot a wrong assumption, a hallucination, or a path that will lead nowhere, you can Ctrl+C immediately — before Claude writes files, runs commands, or burns through tokens on a dead end.

Update (v2.1.x): since a recent update, on a subscription the thinking comes redacted by default, so Ctrl+O alone is no longer enough: you also need "showThinkingSummaries": true in your settings.json. The full, tested detail is in how to bring back Claude Code's hidden thinking.

Result:

> Ctrl+O

Verbose output ON

⏺ Thinking…
  Let me analyze the authentication flow. The user wants to add
  JWT refresh tokens. I'll need to modify the middleware first,
  then update the token service...

  [thinking continues in real-time as gray italic text]

How to use it

1. Toggle verbose output on

Press Ctrl+O at any point during a session. A confirmation message appears and all subsequent responses include the thinking trace. (If you see nothing, remember the note above: on a subscription you need showThinkingSummaries: true.)

2. Watch the reasoning in real time

As Claude streams its response, you'll see the thinking blocks rendered in gray italic text above the actual output. This is the model's internal reasoning — not a summary, the actual chain of thought.

3. Interrupt when needed

If the reasoning reveals a wrong direction, press Ctrl+C to stop generation immediately. Then redirect with a corrected prompt. For when to stop cold with Esc versus steer on the fly, see stopping and redirecting Claude mid-task.

On Fable 5 you also get a warning that doesn't depend on Ctrl+O at all: the (summarized) line tells you what it is about to do right before the tools run.

4. Toggle off when done

Press Ctrl+O again to return to clean output. Verbose mode persists during the session but resets when you exit.

Reference

That last shortcut comes with a catch: Alt+T/Option+T does nothing on Fable 5, where thinking can't be turned off at all. Which control is still wired up on which model is in you're turning knobs that aren't connected to anything.

When verbose mode is most useful

Official docs: Interactive mode — Keyboard shortcuts