Claude Code ultrathink: The Mysterious Prompt for AI Deep Thinking Mode

What is ultrathink?

In the Claude Code system, there exists a "prompt" — **ultrathink**. This prompt can trigger Claude's "extended thinking mode," allocating additional computation time and thinking budget for the AI.

Thinking Level System

In the Claude system, thinking modes are divided into the following levels:

"think" < "think hard" < "think harder" < "ultrathink"

Each level allocates progressively more thinking budget for Claude, allowing the AI to evaluate various solutions more thoroughly.

Officially Recognized Method

This is not a trick discovered by the community, but an officially recognized feature by Anthropic. According to the Claude Code Best Practices official documentation:

Ask Claude to make a plan for how to approach a specific problem. We recommend using the word "think" to trigger extended thinking mode, which gives Claude additional computation time to evaluate alternatives more thoroughly. These specific phrases are mapped directly to increasing levels of thinking budget in the system: "think" < "think hard" < "think harder" < "ultrathink." Each level allocates progressively more thinking budget for Claude to use.

How It Works

"ultrathink" or "think hard" and similar prompts trigger Claude Code's "extended thinking mode." This means the system allocates additional computation time or "thinking budget" for Claude, enabling it to evaluate various solutions more thoroughly.

Technical Principle

This behavior is likely a hard-coded feature. When Claude Code detects specific strings (such as "ultrathink"), it directly sets a specific thinking token budget. This means it's not necessarily an inherent deep or sequential thinking capability of the model itself, but more like a direct instruction that triggers a programmatic response preset in the system related to thinking resource allocation.

In other words, these prompts are "explicitly commanding" the model to invest more resources in thinking.

Recommended Usage Scenarios

You can try using the "ultrathink" prompt when encountering the following situations:

• AI gets stuck in a loop, repeating the same mistakes
• AI keeps making errors and cannot find the correct solution
• It feels like AI has forgotten what it's doing
• Need to handle complex logical problems
• Encountering code problems that require deep analysis

Community Discovery and Verification

This feature was initially discussed and shared by the community on Hacker News, and has been verified through actual testing to be indeed effective. Many developers have reported that after using ultrathink, Claude Code's performance in handling complex problems has significantly improved.

Claude Code ultrathink References

• Mastering Claude Code in 30 minutes - Anthropic YouTube
• Claude Code Best Practices - Anthropic
• Ultrathink is a Claude Code magic word - Hacker News

Claude Code ultrathink Summary

In the Claude Code system, thinking modes are divided into the following levels: "think" < "think hard" < "think harder" < "ultrathink"

Claude Code ultrathink is a built-in powerful feature that can provide additional thinking resources when encountering complex programming problems. Although it increases token consumption, when used at critical moments, it can significantly improve the success rate of solving programming problems.