What Is Thinking (Extended Thinking)?
Claude’s Thinking (Extended Thinking / Extended Reasoning) is a feature for composing answers to complex problems by reviewing them more deeply than usual. As of 2025, it is especially effective for tasks like difficult reasoning, multi-step decision-making, and design exploration.
In intuitive terms, normal mode is like “answer quickly first,” whereas Thinking is “take a bit more time and answer while organizing assumptions, comparing options, and verifying.” The key is not necessarily to use it for every question, but to use it in situations where the thinking process itself directly determines the quality.
- You want to organize problems with many conditions
- You want to produce design proposals from ambiguous requirements
- You want to compare multiple options and come up with the best one
- You want to reduce logical leaps
On the other hand, tasks like short summaries, email drafting, and simple translations are faster and sufficiently practical in normal mode.
How to Enable It and Basic Usage
In Claude’s UI or supported plans, there may be a toggle for Thinking / Extended Thinking within model selection or response settings. Names and placement can change over updates, but the overall flow is generally the same.
- Start a conversation in Claude
- Select the model to use
- Enable Thinking in the settings panel or near the input field
- Ask for complex tasks with conditions specified
Enabling it alone doesn’t automatically improve accuracy. What matters is clearly stating “what to compare, under what constraints, and what you want as the final output.”
You are a senior architect.
Based on the requirements below, compare three proposed architectures for a Web application.
Requirements:
- 10,000 initial users, scale up to 100,000 in the future
- Monthly infrastructure budget within 200,000 yen
- Has an admin dashboard
- Handles personal information
- Development team consists of 3 people
Output format:
1. Overview of each architecture proposal
2. Advantages and disadvantages
3. Estimated costs
4. Recommended proposal and reasonsBy providing role, requirements, constraints, and output format as a set like this, Thinking’s strengths are more likely to come through.
Which Types of Tasks It’s Effective For
1. Mathematics and quantitative problems
It works well for problems that require tasks like formula transformations, conditional branching, and checking multiple cases. Instead of only requesting “the answer,” specifying what should be checked along the way can reduce mistakes.
