Choosing the right model
A simple decision process beats memorizing specs. Here's how to choose in seconds.
The 3-question test
- Is this hard reasoning or high-stakes? (complex analysis, important writing, tricky code, big decisions) → Use Opus (or a Mythos-class model like Fable for the very hardest, long-running work).
- Is this everyday work of normal difficulty? (drafting, summarizing, routine analysis, most coding) → Use Sonnet: the sensible default.
- Is this simple, repetitive, or high-volume, and speed matters? (classification, quick rewrites, bulk processing) → Use Haiku.
Default to Sonnet
When unsure, start with Sonnet. If the answer feels shallow or it makes reasoning mistakes, escalate to Opus. If it's slow and the task is trivial, drop to Haiku.
A quick reference
| Task | Recommended start |
|---|---|
| Draft an email or post | Sonnet (Haiku if you're in a hurry) |
| Summarize a long document | Sonnet |
| Analyze a messy dataset & reason about it | Opus |
| Brainstorm dozens of quick ideas | Haiku / Sonnet |
| Write a board-ready strategy memo | Opus |
| Triage/label hundreds of items | Haiku |
| Multi-hour autonomous coding/research agent | Fable (Mythos) or Opus |
| Production code you ship today | Opus |
Cost vs. quality (for builders)
If you use the API, the trade-off is concrete. Heavier tiers cost more per million tokens. Rough mid-2026 ordering:
text
Haiku < Sonnet < Opus < Fable (Mythos)
cheap premium (~2× Opus)Smart patterns to control cost:
- Tiered pipelines. Do bulk work with Haiku/Sonnet, and escalate only the hard turns to Opus or Fable.
- Prompt caching. Reusing a large fixed context (like a long document or system prompt) can cut input costs dramatically.
- Right-size output. Ask for concise answers when you don't need an essay.
Don't over-optimize
For most individual users on a subscription, the honest advice is: use Sonnet for everyday work and Opus for the hard stuff. Switching models is a smaller lever than writing a better prompt: that's where most quality gains actually come from.
Next: how to spend "thinking" wisely with thinking & effort.