Extended Thinking for Hard Problems

When to Use It

Architecture decisions, tricky debugging, math-heavy tasks and anything where the first answer is usually wrong. Extended thinking buys reasoning effort where it pays off.

Practical Setup

response = client.messages.create(
    model="claude-sonnet-4-5-20250929",
    max_tokens=16000,
    thinking={"type": "enabled", "budget_tokens": 10000},
    messages=[{"role": "user", "content": problem}]
)

Tips

Set the thinking budget to roughly half of max tokens. Ask Claude to show alternatives it considered, not just the winner. For routine tasks leave thinking off: it costs time and money per token.