Extended Thinking
Give Claude a reasoning scratchpad before it answers: thinking budgets, thinking blocks with tool use, interleaved thinking, and when the cost/quality trade-off is worth it.
Last updated
Extended Thinking: Reasoning on Demand
Some problems need more than a fast answer — they need the model to work through them. Extended thinking gives Claude a scratchpad it reasons in before replying, trading tokens and latency for a real jump in quality on hard, multi-step tasks.
By default Claude answers in one pass — great for most things, thin for genuinely hard reasoning. Extended thinking turns on an explicit reasoning phase: you give Claude a thinking budget, it produces internal thinking blocks where it plans, explores, and checks itself, and then writes the final answer conditioned on that work. It’s the difference between blurting and deliberating. The catch is symmetrical: you pay for every thinking token and wait for them, so extended thinking is a dial you turn up for hard problems and leave off for easy ones — not a global setting.
Related
More in LLM & Agentic
LLM Lifecycle
PreviewComplete lifecycle of large language models from pre-training through fine-tuning, RLHF, and deployment — with architecture diagrams and production considerations.
How LLMs Call Tools
PreviewHow LLMs use function calling and tool use — the mechanics behind tool-calling agents, from prompt engineering to structured output.
When to Fine-Tune: The Decision Framework
PreviewShould you fine-tune at all? A structured decision framework for prompt engineering vs RAG vs fine-tuning.
LoRA, QLoRA & PEFT Methods
PreviewHow LoRA works inside transformer layers, QLoRA for memory-efficient training, and the full PEFT method comparison with code examples.
Get full access to all 87+ sections with code examples, diagrams, and interactive animations.
Unlock Premium