Claude Opus 4.7 vs. Sonnet 4.6 — which Claude to pick

Open the Anthropic model picker and you will see Opus 4.7 at the top and Sonnet 4.6 right under it. Both have a 1 million token context window, both handle vision, both run extended thinking. So why is Opus five times more expensive per output token? The honest answer: Opus earns that premium on a small slice of tasks. For the other 80% of your work, Sonnet is the right call.

Reach for Opus when

• You need to refactor code that spans 5+ files and depends on subtle invariants
• You are drafting a legal brief, policy memo, or regulatory filing where mistakes cost real money
• You are pushing a long-horizon agent (Claude Code on a complex branch, auto-PR generation)
• You are analyzing a 500-page report and the answer matters for a real decision

Sonnet is enough when

• Customer-support chatbot answering product questions
• Daily email drafts, meeting notes, summaries
• Code review on single files or small PRs
• Routine document Q&A where citations matter more than depth

from anthropic import Anthropic

client = Anthropic()

# Opus for a hard coding job
opus_reply = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    thinking={"type": "enabled", "budget_tokens": 8000},
    messages=[{"role": "user", "content": "Refactor this multi-file auth module..."}],
)

# Sonnet for everyday chat — same shape, cheaper
sonnet_reply = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Summarize this meeting transcript."}],
)