Extended Thinking
Deep reasoning mode that lets Claude think through complex problems before responding.
Extended Thinking allows Claude to reason through complex problems before responding. When enabled, Claude's internal reasoning process is displayed in a collapsible section above the final response.
How It Works
With Extended Thinking enabled:
- Claude receives your message and begins an internal reasoning phase
- The reasoning appears in real time in a collapsible "Thinking" section
- After reasoning, Claude produces a more thorough, well-considered response
- The thinking tokens are counted separately from the response tokens
The thinking process is especially valuable for complex tasks where Claude needs to consider multiple approaches, understand existing code patterns, or reason about edge cases.
Configuration
Enable Extended Thinking
Go to Settings (Cmd+,) > AI > Extended Thinking and toggle it on.
Set the Thinking Budget
Configure Max Thinking Tokens to control how much reasoning Claude can do. The default is 16,000 tokens. Higher values allow deeper reasoning but increase cost and latency.
When to Use Extended Thinking
Extended Thinking adds the most value for:
- Complex refactoring — Understanding interrelated code before making changes
- Architecture decisions — Weighing trade-offs between approaches
- Unfamiliar codebases — Building a mental model of existing patterns
- Bug diagnosis — Reasoning through potential causes systematically
- Multi-file changes — Planning coordinated modifications
Extended Thinking pairs well with Plan Mode. Enable both to let Claude think deeply about the problem, research the codebase in read-only mode, and present a well-reasoned plan for your approval.
Thinking Budget Tips
- Start with the default (16,000 tokens) for most tasks
- Increase to 32,000+ for complex architectural work
- Lower to 8,000 for simpler tasks where you want faster responses
- The thinking budget is part of your overall cost controls
Viewing Thinking Output
During streaming, the thinking section appears as a collapsible block above Claude's response. Click to expand or collapse it. The thinking content shows Claude's step-by-step reasoning, including what it considered and why it chose a particular approach.