Mastering Extended Thinking in Claude: A Complete Guide to Adaptive & Manual Thinking Modes

Claude's extended thinking feature unlocks deeper reasoning capabilities for complex tasks. Whether you're building a multi-step analysis tool, a code review assistant, or a research aid, extended thinking gives Claude the space to "think through" problems before delivering its final answer—and optionally shows you its step-by-step reasoning.

This guide covers everything you need to know: how extended thinking works, which modes to use for each model, how to configure it in the API, and practical tips for getting the most out of it.

What Is Extended Thinking?

When extended thinking is enabled, Claude creates internal thinking content blocks where it performs step-by-step reasoning. It then uses the insights from that reasoning to craft its final response. The API response includes both the thinking blocks (with varying levels of transparency) and the final text blocks.

This is especially useful for:

• Complex mathematical or logical problems
• Multi-step reasoning tasks (e.g., legal analysis, code debugging)
• Tasks requiring careful planning before execution
• Any scenario where you want to see how Claude arrived at its answer

Adaptive Thinking vs. Manual Extended Thinking

Claude now offers two approaches to extended thinking. The right choice depends on your model and use case.

Adaptive Thinking (Recommended)

Adaptive thinking (thinking: {type: "adaptive"}) lets Claude dynamically decide how much thinking to do based on the complexity of the task. You can optionally control the effort level. This is the modern, recommended approach. Supported on: Claude Opus 4.7, Claude Opus 4.6, Claude Sonnet 4.6, and Claude Mythos Preview.

Manual Extended Thinking (Deprecated)

Manual extended thinking (thinking: {type: "enabled", budget_tokens: N}) lets you set a fixed token budget for thinking. This approach is deprecated and will be removed in future model releases. It is no longer accepted on Claude Opus 4.7 (returns a 400 error). Still functional on: Claude Opus 4.6, Claude Sonnet 4.6, and Claude Mythos Preview (but deprecated).

Model-Specific Behavior

How to Use Extended Thinking in the API

1. Adaptive Thinking with Effort (Claude Opus 4.7)

For Claude Opus 4.7, use adaptive thinking with the effort parameter. Effort controls how much reasoning Claude applies—higher effort means deeper thinking but longer response times.

Python Example:

import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    thinking={
        "type": "adaptive",
        "effort": "high"  # Options: "low", "medium", "high"
    },
    messages=[
        {"role": "user", "content": "Analyze the logical consistency of this argument: All men are mortal. Socrates is a man. Therefore, Socrates is immortal."}
    ]
)
Print the thinking content
for block in response.content:
    if block.type == "thinking":
        print("Thinking:", block.thinking)
    elif block.type == "text":
        print("Final answer:", block.text)

TypeScript Example:

import Anthropic from '@anthropic-ai/sdk';
const client = new Anthropic();
async function main() {
  const response = await client.messages.create({
    model: 'claude-opus-4-7',
    max_tokens: 4096,
    thinking: {
      type: 'adaptive',
      effort: 'high'
    },
    messages: [
      { role: 'user', content: 'Analyze the logical consistency of this argument: All men are mortal. Socrates is a man. Therefore, Socrates is immortal.' }
    ]
  });
for (const block of response.content) {
    if (block.type === 'thinking') {
      console.log('Thinking:', block.thinking);
    } else if (block.type === 'text') {
      console.log('Final answer:', block.text);
    }
  }
}
main();

2. Manual Extended Thinking (Deprecated, Older Models)

If you're using Claude Opus 4.6 or Sonnet 4.6 and haven't migrated yet, you can still set a manual token budget:

response = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=8192,
    thinking={
        "type": "enabled",
        "budget_tokens": 4096  # Max tokens for thinking
    },
    messages=[
        {"role": "user", "content": "Solve this complex math problem step by step..."}
    ]
)

Important: The budget_tokens value must be less than max_tokens. It sets the upper limit on how many tokens Claude can use for internal reasoning.

3. Task Budgets (Beta)

For advanced use cases, you can set a task budget—a target for total tokens (thinking + output). This helps control costs and latency while still allowing adaptive thinking.

response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=8192,
    thinking={
        "type": "adaptive",
        "effort": "medium",
        "task_budget_tokens": 4096  # Beta: target total tokens
    },
    messages=[
        {"role": "user", "content": "Write a detailed analysis of quantum computing..."}
    ]
)

4. Fast Mode (Research Preview)

Fast mode reduces thinking time for simpler tasks while still using extended thinking. This is useful when you want reasoning but need faster responses.

response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    thinking={
        "type": "adaptive",
        "effort": "low",
        "fast_mode": True  # Research preview
    },
    messages=[
        {"role": "user", "content": "Summarize this article in 3 bullet points..."}
    ]
)

Understanding the Response Format

When extended thinking is enabled, the API response includes thinking content blocks before the final text block:

{
  "content": [
    {
      "type": "thinking",
      "thinking": "Let me analyze this step by step...",
      "signature": "WaUjzkypQ2mUEVM36O2TxuC06KN8xyfbJwyem2dw3URve/op91XWHOEBLLqIOMfFG/UvLEczmEsUjavL...."
    },
    {
      "type": "text",
      "text": "Based on my analysis..."
    }
  ]
}

The signature field is used internally for verification and can be ignored in most applications.

Best Practices

• Use adaptive thinking for new projects. Manual mode is deprecated and will break on future models.
• Match effort to task complexity. Use "low" for simple Q&A, "medium" for standard analysis, and "high" for complex reasoning.
• Set max_tokens generously. Extended thinking consumes tokens internally; ensure your max_tokens is at least 2x your expected output length.
• Handle thinking blocks in streaming. If streaming, thinking blocks appear as content_block_delta events with type thinking_delta.
• Test with task budgets (beta). If cost or latency is a concern, experiment with task_budget_tokens to find the sweet spot.

Differences Across Model Versions

Thinking behavior varies by model version. Key differences:

• Claude Opus 4.7: Only supports adaptive thinking; manual mode returns 400 error.
• Claude Opus 4.6 & Sonnet 4.6: Adaptive recommended; manual still works but deprecated.
• Claude Mythos Preview: Adaptive is default; disabled not supported; use display: "summarized" for summarized thinking.

Always check the model documentation for the latest behavior.

Key Takeaways

• Extended thinking gives Claude enhanced reasoning capabilities with transparent step-by-step thinking.
• Adaptive thinking (type: "adaptive") is the recommended approach for all current models, especially Claude Opus 4.7.
• Manual extended thinking (type: "enabled" with budget_tokens) is deprecated and will be removed in future releases.
• Use the effort parameter (low/medium/high) to control reasoning depth without manual token budgets.
• Task budgets (beta) and fast mode (research preview) offer additional control over cost and latency.
• Always set max_tokens higher than your expected output to accommodate internal thinking tokens.