Haiku, Sonnet, Opus or Fable?
The most powerful model isn't always the best one for the job. Running every short email through Fable 5 means paying too much and waiting too long. Test yourself: cover the answer and guess along.
Reading time ± 7 minutes · 36-prompt quiz
Why model choice matters
The names hint at it. A haiku is shorter than a sonnet, an opus outweighs them both — and since July 2026 there's one more story above that: Fable, the Mythos-class model. Choose based on what the task actually needs, not on what sounds most powerful.
Below are 36 concrete prompts in random order. For each one, guess which model you'd use. Click your choice and you'll see right away whether it's right — plus your score at the bottom.
The four models in brief
Four tiers, four roles. Once these click, most prompts answer themselves.
The sprinter: fast, cheap, high volume.
Best for: Short, clear tasks at scale: classifying, tagging, extracting, summarizing and standard replies.
The all-rounder: your default choice.
Best for: Everyday work: writing, coding, debugging, analyzing and brainstorming. The best balance of speed and intelligence.
The heavyweight: for complex tasks.
Best for: Heavy lifting: deep reasoning, architecture, multi-step agents and scenario-based strategy.
The flagship: Mythos-class above Opus.
Best for: Your very toughest challenges: multi-day autonomous agent work, huge migrations and the deepest research and reasoning.
The quiz: which model would you use?
Read the prompt, pick your model and get instant feedback. No account, no fuss.
- 1
“Understand this entire codebase and draw up a refactor plan.”
- 2
“Summarize this news article in 3 sentences.”
- 3
“Build a React component for a price comparison tool.”
- 4
“Work out a complex legal question in multiple steps.”
- 5
“Tag these reviews as positive, neutral or negative.”
- 6
“Think along about my content strategy for Q3.”
- 7
“Devise a complete go-to-market strategy with scenarios.”
- 8
“Sort these 200 support messages by topic.”
- 9
“Analyze this contract and pull out the key risks.”
- 10
“Design the architecture for a new data platform.”
- 11
“Correct the spelling mistakes in this message.”
- 12
“Write a 600-word blog post about home batteries.”
- 13
“Plan a migration of a large application with zero downtime.”
- 14
“Give a short, friendly acknowledgement of receipt.”
- 15
“Rewrite this landing page to be more conversion-focused.”
- 16
“Solve this deep reasoning puzzle and show your reasoning.”
- 17
“Extract all email addresses from this text.”
- 18
“Debug this function and explain what went wrong.”
- 19
“Build a multi-step agent that fetches, analyzes and reports on data.”
- 20
“Turn this bullet list into clean sentences.”
- 21
“Make a structured summary of this 20-page report.”
- 22
“Analyze 3 competitor strategies and build a well-supported recommendation.”
- 23
“Answer this frequently asked customer question.”
- 24
“Draft 10 interview questions for a marketing role.”
- 25
“Produce a scientific literature review on a difficult topic.”
- 26
“Translate this standard sentence into German.”
- 27
“Write a 4-email onboarding sequence.”
- 28
“Write and optimize a complex script with edge cases.”
- 29
“Turn this product description into a one-liner.”
- 30
“Create a product description plus 3 ad variants.”
- 31
“Migrate our entire legacy monolith to a new framework, working on it autonomously for days.”
- 32
“Turn this meeting note into an action list with owners.”
- 33
“Run full due diligence on this acquisition file: hundreds of documents, one substantiated final report.”
- 34
“Find the root cause of this stubborn production bug three teams have been stuck on for weeks.”
- 35
“Write the onboarding email sequence for our new customers (5 emails).”
- 36
“Set up a multi-day autonomous agent that modernizes our entire test suite, working pull requests included.”
The rule of thumb
Start with Sonnet. Scale down to Haiku when speed and volume matter. Scale up to Opus when the task gets genuinely complex. And save Claude Fable 5 — the Mythos-class model above Opus — for your very toughest challenges: multi-day autonomous tasks and the deepest reasoning work.
Want to work smarter with Claude Cowork?
Browse the skills library: ready-to-run workflows you can paste straight into Claude.
Go to the skills libraryWhich model do you reach for by default?