Output styles — Tune verbosity

What it does

Output styles control how verbose Claude's responses are. Claude Code ships with three presets — terse (one-line answers to direct questions), balanced (default: explanations when needed, brief when not), and verbose (full reasoning and walkthrough). You can also create custom styles that dial in exactly how much detail you want, and switch between them mid-session without restarting — one message terse, the next verbose, whatever fits the task.

When to use it

Use terse when you want quick answers: a one-liner explaining what a file does, or "yes, that's the right approach." Use balanced (the default) for most work — it adapts: short replies for straightforward questions, longer ones when the task is complex or you're learning something new. Use verbose when you're debugging a tricky issue and want to see the agent's full reasoning, or when you're being taught and want comprehensive explanations with examples.

Mid-session switching is powerful. You might start a feature build in balanced mode (efficient), hit a confusing bug, switch to verbose to see the diagnostic reasoning, then flip back to terse once you understand the problem and just want quick answers while fixing it.

Try it yourself

Open /config in Claude Code and navigate to the output style setting — you'll see the presets and an option to create a custom style by picking a base tone (minimal, concise, conversational, detailed) and optional tweaks for verbosity, examples, and reasoning transparency. Try switching between terse and verbose on the same type of question (e.g., "is this code correct?") in two separate messages to feel the difference. Once you find a preference, set it as your default; if you need something different for a specific task, type /config again mid-session to change it.

Gotchas

Output styles apply only to new responses — they don't retroactively reformat messages already sent, so don't expect to switch styles and see earlier answers condense. Styles also don't affect tool output (file contents, git diff, test results) — only Claude's prose around those outputs. If you create a custom style and save it, it's remembered across sessions, but if you flip between presets rapidly ("terse for this one, verbose for that one"), the last one you pick sticks until you change it again — easy to forget you're in terse mode and wonder why Claude isn't giving you explanations. Finally, verbose mode can feel slow if you're used to terse; it's not actually slower (same model, same latency), just longer to read, so don't mistake verbosity for lag.