Skip to content
Last Dictionary
Plane
Language Models & TransformersAI Engineering & MLOps
emerging

extended thinking

nounAlso: thinking mode, thinking blocks, adaptive thinking

In plain wordsA setting in some AI models, notably Anthropic's Claude, that lets the model write out private step-by-step reasoning before its final answer, improving results on hard problems.

  1. Anthropic's name for the reasoning mode of its Claude models, in which the model emits 'thinking' content blocks — its reasoning tracereasoning trace, n.: The written-out 'thinking' an AI reasoning model produces before its final answer — notes to itself full of attempts, checks and corrections, which users may or may not get to see. — before the visible answer. As first offered with Claude 3.7 Sonnet in February 2025, developers enabled it per request with a token budget (minimum 1,024 tokens), and thinking tokens were billed as output. Later models added interleaved thinkinginterleaved thinking, n.: A way for an AI agent to pause and reason again after every action it takes — reading each search result or test output before deciding what to do next — instead of only thinking once at the start. between tool callstool call, n.: A single request from an AI model to use one of its tools — naming the tool and filling in the details — which the surrounding software then carries out and reports back on., returned summarized rather than raw thinking, and replaced fixed budgets with adaptive thinking, in which the model decides whether and how much to think under a coarse effort setting; as of October 2026 Anthropic's documentation treats manual budgets as legacy and rejects them on its newest models. The phrase is also used generically for extended test-time reasoningtest-time compute, n.: The computing effort an AI model spends while answering a question, as opposed to while being trained. Letting a model think longer or try many answers can make it markedly smarter. by any reasoning modelreasoning model, n.: An AI model trained to work through a problem step by step in hidden 'thinking' text before it answers. Letting it think longer generally makes it better at math, coding and planning..

    Turn on extended thinking for the migration plan, but not for the chat replies.

Origin

Introduced by Anthropic with Claude 3.7 Sonnet on 24 February 2025, which could answer quickly or, with extended thinking enabled, reason at length. From the Claude 4.6 generation onward, adaptive thinking and effort levels superseded fixed thinking budgets.

Related

See also
  • reasoning modelreasoning model, n.: An AI model trained to work through a problem step by step in hidden 'thinking' text before it answers. Letting it think longer generally makes it better at math, coding and planning.
  • thinking budgetthinking budget, n.: A limit, set by whoever uses an AI model, on how much private step-by-step reasoning it may do before answering: more thinking usually means better answers but higher cost and slower replies.
  • interleaved thinkinginterleaved thinking, n.: A way for an AI agent to pause and reason again after every action it takes — reading each search result or test output before deciding what to do next — instead of only thinking once at the start.
  • hybrid reasoning modelhybrid reasoning model, n.: A single AI model that can either answer right away or first think at length, with the user or developer choosing how much thinking to allow.
  • reasoning tracereasoning trace, n.: The written-out 'thinking' an AI reasoning model produces before its final answer — notes to itself full of attempts, checks and corrections, which users may or may not get to see.
  • test-time computetest-time compute, n.: The computing effort an AI model spends while answering a question, as opposed to while being trained. Letting a model think longer or try many answers can make it markedly smarter.

Sources

  1. Anthropic (2026). Extended thinking. Claude Developer Platform documentation. platform.claude.com/docs/en/build-with-claude/extended-thinking
  2. Anthropic (2026). Adaptive thinking. Claude Developer Platform documentation. platform.claude.com/docs/en/build-with-claude/adaptive-thinking

Mentioned in

ClaudeClaude, prop. n.: A family of AI assistants and language models made by the company Anthropic since 2023, widely used for writing, analysis and computer programming., reasoning modelreasoning model, n.: An AI model trained to work through a problem step by step in hidden 'thinking' text before it answers. Letting it think longer generally makes it better at math, coding and planning., QwenQwen, prop. n.: A family of AI language models from the Chinese company Alibaba, many free to download, which by 2025 had become the most popular starting point for other people's open AI models., reasoning tracereasoning trace, n.: The written-out 'thinking' an AI reasoning model produces before its final answer — notes to itself full of attempts, checks and corrections, which users may or may not get to see., thinking budgetthinking budget, n.: A limit, set by whoever uses an AI model, on how much private step-by-step reasoning it may do before answering: more thinking usually means better answers but higher cost and slower replies., hybrid reasoning modelhybrid reasoning model, n.: A single AI model that can either answer right away or first think at length, with the user or developer choosing how much thinking to allow., interleaved thinkinginterleaved thinking, n.: A way for an AI agent to pause and reason again after every action it takes — reading each search result or test output before deciding what to do next — instead of only thinking once at the start.