Correction: Verbose Output Is Not Extended Thinking
Published: December 8, 2025 • 3 min read
This is going to be a short one because its aim is to correct claims I made in two previous blog posts.
In Claude God Tip #10, I claimed that the "verbose output" feature in Claude Code is the equivalent of the "Extended Thinking" feature on Claude's web interface. Then in my post about Extended Thinking tokens, I admitted I was starting to question myself.
Now I can say with confidence: I was wrong.
They are different features, especially when you consider them from a token perspective.
What Extended Thinking Does (Claude on the Web)
Extended Thinking on the web:
- Allocates a separate "thinking budget" of tokens
- Claude explicitly reasons through the problem before responding
- You see this reasoning in an expandable dropdown
- These thinking tokens are billed but don't persist in context
What Verbose Output Does (Claude Code)
Verbose output in Claude Code:
- Shows you the tool calls Claude is making (file reads, edits, bash commands)
- Displays internal planning and decision-making text
- It's a visibility setting, not a reasoning mode toggle
The Key Difference
Extended Thinking gives Claude dedicated tokens to "think before speaking." Verbose output simply shows you what Claude Code is already doing behind the scenes.
The ultimate takeaway: when using Extended Thinking with Claude on the web, you get deeper reasoning without bloating your context window because the thinking tokens are separate and temporary.
Claude Code's verbose mode is useful for understanding what actions the built-in general purpose subagent is taking, but it's a different mechanism entirely.
I hope this clarifies any confusion that may have come from the previous blog posts.
As always, thanks for reading!