Claude Extended Thinking API — Python Examples

How to use Claude's extended thinking mode in Python. Working code for streaming and non-streaming extended thinking with token budgets and cost estimates.

Claude's extended thinking mode lets the model reason through hard problems before answering. This guide shows how to enable it in Python, interpret the response, and control costs.

Basic extended thinking call

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=16000,  # must be > budget_tokens
    thinking={
        "type": "enabled",
        "budget_tokens": 10000  # how many tokens Claude can use for reasoning
    },
    messages=[{
        "role": "user",
        "content": "What is the most efficient algorithm for finding the k-th largest element in an unsorted array? Compare time and space complexity of at least three approaches."
    }]
)

# Response has two content blocks: thinking + text
for block in response.content:
    if block.type == "thinking":
        print("=== Claude's Reasoning ===")
        print(block.thinking[:500], "...")  # show first 500 chars
    elif block.type == "text":
        print("\n=== Final Answer ===")
        print(block.text)

# Token usage
print(f"\nInput tokens: {response.usage.input_tokens}")
print(f"Output tokens (incl. thinking): {response.usage.output_tokens}")

Streaming extended thinking

with client.messages.stream(
    model="claude-sonnet-4-6",
    max_tokens=16000,
    thinking={"type": "enabled", "budget_tokens": 8000},
    messages=[{"role": "user", "content": "Debug this recursive function and explain why it has exponential time complexity:\n\ndef fib(n):\n    if n <= 1: return n\n    return fib(n-1) + fib(n-2)"}]
) as stream:
    current_block_type = None
    for event in stream:
        if hasattr(event, "type"):
            if event.type == "content_block_start":
                current_block_type = event.content_block.type
                if current_block_type == "thinking":
                    print("\n[Thinking...]", end="", flush=True)
                elif current_block_type == "text":
                    print("\n\n[Answer]\n", end="", flush=True)
            elif event.type == "content_block_delta":
                if hasattr(event.delta, "thinking"):
                    print(".", end="", flush=True)  # progress dots for thinking
                elif hasattr(event.delta, "text"):
                    print(event.delta.text, end="", flush=True)

Choose the right thinking budget

# Thinking budget guidelines
BUDGETS = {
    "quick_reasoning": 1000,    # simple logic, short proofs
    "standard_reasoning": 5000, # coding problems, analysis
    "deep_reasoning": 16000,    # complex math, research tasks
    "maximum": 32000,           # hardest problems, long-form planning
}

# Cost estimate per call (Claude Sonnet at $15/M output tokens)
for label, budget in BUDGETS.items():
    thinking_cost = (budget / 1_000_000) * 15  # thinking billed as output
    print(f"{label:25s}  budget={budget:6d}  max thinking cost=${thinking_cost:.4f}")

Extended thinking for multi-step coding tasks

def deep_code_review(code: str, language: str = "python") -> dict:
    response = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=16000,
        thinking={"type": "enabled", "budget_tokens": 8000},
        messages=[{
            "role": "user",
            "content": f"""Perform a deep code review of this {language} code. Think through:
- Time and space complexity
- Edge cases and potential bugs
- Security vulnerabilities
- Refactoring opportunities

Code:
{code}

Provide a structured review with severity ratings."""
        }]
    )

    thinking = next((b.thinking for b in response.content if b.type == "thinking"), "")
    answer = next((b.text for b in response.content if b.type == "text"), "")

    return {
        "reasoning": thinking,
        "review": answer,
        "tokens_used": response.usage.output_tokens,
        "cost_usd": (response.usage.output_tokens / 1_000_000) * 15
    }

Cost estimate by thinking budget

Calculate costs for your specific thinking workloads with the Claude API Cost Calculator.

Frequently asked questions

Free tools

More examples