ClaudeModels & limitsLevel 1: Basic prompts

Which Claude Model Should You Use?

Claude comes in four models: Fable, Opus, Sonnet and Haiku. Start with Sonnet for everyday work, and move up only when you can see the answer falling short. Move down to Haiku when you run the same simple task many times.

Prompt to copy
Here is a task I plan to run [once / every day / thousands of times]:

[describe the task and paste one real example]

The available models are Haiku (fastest, cheapest), Sonnet (balanced), Opus (stronger, slower) and Fable (strongest, slowest).

Recommend the cheapest model likely to do this well. Tell me what I might lose by going one model lower, and give me one test I can run to check.
  • Claude Fable 5.1$10
  • Claude Opus 5.5$4
  • Claude Sonnet 5.5$2
  • Claude Haiku 5.5from $0.10
Input price per million tokens on the API. Source: Anthropic's models overview, 6 November 2026. Haiku's rate applies up to 100,000 prompt tokens.
cost of a wrong answer →
Sonnetsimple, so check it
Opus or Fablehard reasoning, long agent runs
Haikutagging at volume
Sonnetmost everyday writing and coding
how hard the task is →
A rule of thumb, not an Anthropic chart: start in the lower right, and only move when the answer falls short.

Try it now

  1. 1Pick a real task you did this week, and run it in Sonnet.
  2. 2Open a new chat, choose a stronger or a lighter model from the model picker, and run the exact same prompt.
  3. 3Compare the two answers against what you actually needed. If you can't point to a difference, keep the cheaper model.

Use it when

Any time you pick a model in the Claude app, in Claude Code, or in an API call. It matters most for work you repeat, since a poor default gets multiplied every time.

Skip it when

A one-off question. Use the default and move on. If you're on the Free plan you can't choose Opus or Fable anyway, so there's nothing to decide.

Why it works

Stronger models are slower and cost more, and on a plan they use up your limits faster. Anthropic's docs rank the current models from slowest to fastest as Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, and API input prices run from $10 down to $0.10 per million tokens. Testing your own task on two models tells you more than any label does, because many tasks show no visible difference.

Example

Say you need to tag 10,000 support tickets by topic. This is the job Anthropic describes Haiku for: high-volume classification and extraction. At $0.10 per million input tokens, it costs a fortieth of Opus on input. Then, once a week, you ask for a memo on what those tickets say about your product. That one is harder and read by people who'll act on it, so it's worth testing on Sonnet and Opus side by side.

Going further

On the API, use two steps. Haiku handles every item first and flags the ones it's unsure about, and only those go to Opus. Watch one trap: Haiku 5.5's lower rate applies up to 100,000 prompt tokens per request. Above that, its input rate rises from $0.10 to $0.50 per million tokens, so keep requests small or send long ones to Sonnet.

Common mistakes

  • Using the strongest model for everything, then running out of your plan's limits by midweek.
  • Dropping to a small model for hard reasoning and blaming AI for the weak answer.
  • Comparing models on different prompts. Change one thing at a time, or the test tells you nothing.
  • Assuming Fable is included in Pro. Anthropic's plan page says Fable on Pro runs on usage credits, charged at API rates.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Beyond the course