What is the minimum token budget for Extended Thinking?

The minimum budget required to activate Claude's extended thinking mode is 1,024 tokens.

Which models support this reasoning capability?

Extended Thinking is supported by the Claude 4 and 4.5 model families, including Opus, Sonnet, and Haiku variants.

When should I use streaming with this skill?

Streaming is required when your max_tokens exceed 21,333 and is generally recommended for better UX during long reasoning processes.

Why must I preserve thinking blocks when using tools?

Preserving thinking blocks in the conversation history is critical for multi-turn tool use to ensure Claude maintains its logical context and reasoning chain.

Does Extended Thinking allow for custom temperature settings?

No, when extended thinking is enabled, the temperature must be set to 1 and cannot be modified.

Extended Thinking (Ultrathink)

Name: Extended Thinking (Ultrathink)
Author: Lobbi-Docs

byLobbi-Docs

Herramientas para Desarrolladores

Enables high-fidelity internal reasoning and deliberation for complex problem-solving tasks using Claude's extended thinking mode.

Acerca de

The Extended Thinking skill unlocks Claude’s 'ultrathink' capabilities, allowing the model to perform deep internal deliberation before providing a final response. This is essential for tackling highly complex coding challenges, architectural planning, and multi-step logic problems that require systematic analysis rather than immediate pattern matching. By configuring custom thinking budgets and preserving reasoning blocks during tool use, developers can ensure higher accuracy and more thoughtful outputs for mission-critical tasks.

Características Principales

Optimized triggers for deep analysis, strategic planning, and security audits
Configurable reasoning token budgets ranging from 1,024 to 32,000+
0 GitHub stars
Full support for internal deliberation across Claude 4 and 4.5 model families
Seamless integration with tool-calling workflows via thinking block preservation
Streaming-compatible architecture for long-form reasoning and large outputs

Casos de Uso

Architecting complex microservices with multiple technical trade-offs
Debugging deep-seated concurrency or memory leak issues needing systematic analysis
Conducting thorough security threat modeling and vulnerability assessments

Acerca de

Características Principales

Optimized triggers for deep analysis, strategic planning, and security audits
Configurable reasoning token budgets ranging from 1,024 to 32,000+
0 GitHub stars
Full support for internal deliberation across Claude 4 and 4.5 model families
Seamless integration with tool-calling workflows via thinking block preservation
Streaming-compatible architecture for long-form reasoning and large outputs

Casos de Uso

Architecting complex microservices with multiple technical trade-offs
Debugging deep-seated concurrency or memory leak issues needing systematic analysis
Conducting thorough security threat modeling and vulnerability assessments