Claude Opus 4 API Example (claude-opus-4-7)

Working Python examples for Claude Opus 4 (claude-opus-4-7), Anthropic's most capable model. Learn when Opus beats Sonnet, how to use extended thinking with Opus, and how to manage cost at Opus pricing.

Claude Opus 4 (claude-opus-4-7) is Anthropic's most capable model. Use it when task difficulty justifies the higher cost — most production traffic routes to Sonnet 4.6.

Basic Opus 4 call

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-4-7",   # ← Opus 4
    max_tokens=2048,
    messages=[{
        "role": "user",
        "content": "Design a microservices architecture for a real-time bidding system handling 500K requests/second."
    }]
)

print(response.content[0].text)
print(f"Input tokens:  {response.usage.input_tokens}")
print(f"Output tokens: {response.usage.output_tokens}")

Extended thinking with Opus 4

response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=16000,
    thinking={
        "type": "enabled",
        "budget_tokens": 10000  # Opus reasons for up to 10K tokens before answering
    },
    messages=[{
        "role": "user",
        "content": """
        A snail climbs a 20-foot pole. Each day it climbs 3 feet and slides back 2 feet at night.
        Except on the last day, when it reaches the top, it doesn't slide back.
        How many days does it take?
        Show your full reasoning.
        """
    }]
)

for block in response.content:
    if block.type == "thinking":
        print(f"[Opus thinking — {len(block.thinking)} chars]")
    elif block.type == "text":
        print(block.text)

Model routing: Haiku → Sonnet → Opus

def route_to_model(prompt: str, complexity: str) -> str:
    """Route to the right model based on task complexity."""
    model_map = {
        "simple":  "claude-haiku-4-5-20251001",  # classification, extraction, short tasks
        "medium":  "claude-sonnet-4-6",           # most tasks — best price/quality
        "complex": "claude-opus-4-7",             # deep reasoning, architecture design
    }
    model = model_map.get(complexity, "claude-sonnet-4-6")

    response = client.messages.create(
        model=model,
        max_tokens=4096,
        messages=[{"role": "user", "content": prompt}]
    )
    return response.content[0].text

# Simple: Haiku (fast, cheap)
result = route_to_model("Classify this as spam or not: 'You won a prize!'", "simple")

# Medium: Sonnet (most production use cases)
result = route_to_model("Summarize this 10-page technical document.", "medium")

# Complex: Opus (hard reasoning)
result = route_to_model("Design an optimistic concurrency control system for a distributed database.", "complex")

Opus 4 with prompt caching (reduce cost)

long_document = "..." * 50000  # 50K+ token document

response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=2048,
    system=[{
        "type": "text",
        "text": long_document,
        "cache_control": {"type": "ephemeral"}  # cache the document
    }],
    messages=[{"role": "user", "content": "What are the key risk factors in this document?"}]
)

# First call: full input cost. Subsequent calls: 90% discount on cached tokens.
print(f"Cache write tokens: {response.usage.cache_creation_input_tokens}")
print(f"Cache read tokens:  {response.usage.cache_read_input_tokens}")

When Opus 4 vs Sonnet 4.6

For cost modeling across models, use the Claude API Cost Calculator. For extended thinking deep-dive, see Extended Thinking with Claude. For migration from OpenAI's o1/o3 to Claude Opus extended thinking, see the Claude API Cookbook.

Frequently asked questions

Free tools

More examples