Extended thinking
Also written: Extended thinking mode, Open-ended thinking
AI models show intermediate reasoning steps to solve complex problems.
Extended thinking is a feature that allows AI models to pause and write down their thought process before giving a final answer. This is especially helpful for difficult tasks that require multiple steps or complex logic.
Think of it like a student showing their work on a math problem. Instead of just giving the answer, the AI lays out each step it took to get there. This helps the AI solve the problem more accurately and can also help users understand how the AI arrived at its conclusion.
Extended thinking refers to a capability where an AI model explicitly generates intermediate reasoning steps or a scratchpad of its thought process before producing a final output. This mechanism is employed to decompose complex problems, maintain coherence over longer reasoning chains, and improve accuracy on tasks requiring multi-step deduction or planning. It can be seen as a form of internal deliberation or explicit chain-of-thought generation that is made visible to the user or system.
Recent updates from Google Gemini and Anthropic highlight features for more complex AI reasoning. These advancements allow models to break down problems, show intermediate steps, or dedicate more processing time to difficult tasks, improving overall problem-solving capabilities.
- Gemini API Cookbook: feat: add Gemini 3.8 Live Extended Thinking quickstart and thinking level configuration (#1379)Google · Gemini API Cookbook ·
- Anthropic Economic Index: Insights from Claude 3.7 SonnetAnthropic · Anthropic Research ·
- Anthropic Economic Index: Insights from Claude 3.7 SonnetAnthropic · Anthropic News ·
- Evaluating AI’s ability to perform scientific research tasksOpenAI · OpenAI News ·
- The "think" tool: Enabling Claude to stop and thinkAnthropic · Anthropic Engineering ·
- Claude Opus 4.1Anthropic · Anthropic News ·
- Anthropic Claude
- Anthropic's Claude models use a 'think' tool and recommend 'extended thinking' for better complex reasoning and integration of AI applications.
- Google Gemini
- Google Gemini offers advanced thinking levels and an 'extended thinking' quickstart for its API, allowing users to configure deeper reasoning in notebooks.
- Chain of Thought
- A technique where models generate intermediate reasoning steps
- Scratchpad
- A temporary storage area for intermediate computations or thoughts,
- Self-reflection
- A process where AI reviews its own output or reasoning, which c
The concept of models showing intermediate reasoning steps, often referred to as Chain of Thought, gained prominence with research into improving LLM reasoning abilities. Anthropic has specifically termed a feature for this 'extended thinking'. Source ↗
When using a capable AI like Claude, try asking it to solve a multi-step math problem or a logic puzzle. You can often prompt it to 'think step-by-step' or 'show your work' to observe its extended thinking process in action.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- Does extended thinking make the AI slower?
- Yes, generating intermediate reasoning steps typically takes more time and computational resources than providing a direct answer, potentially slowing down the response.
- Can extended thinking be turned off?
- Often, yes. Depending on the AI model and interface, users may have options to enable or disable features that encourage or require extended thinking.
First seen on Subvolts Aug 14, 2025. Updated Oct 5, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.