---
title: "Which Claude Model to Use: Haiku, Sonnet, Opus or Fable"
description: "Which Claude model fits your task? Haiku, Sonnet, Opus and Fable compared, plus a simple rule of thumb you can test on your own work."
author: "Matthias Meyer"
published: 2026-07-09
updated: 2026-09-30
updates:
  - date: 2026-09-30
    note: "Removed unsupported blind tests and percentages; made model and reasoning choices task-specific."
language: en
tags: ["claude", "modelle", "opus", "sonnet", "haiku", "einsteiger", "modell-vergleich"]
canonical: "https://studiomeyer.io/en/blog/which-claude-model"
markdown_versions: ["https://studiomeyer.io/de/blog/which-claude-model.md", "https://studiomeyer.io/en/blog/which-claude-model.md", "https://studiomeyer.io/es/blog/which-claude-model.md"]
publisher: "StudioMeyer, https://studiomeyer.io (llms.txt: https://studiomeyer.io/llms.txt)"
---

# Which Claude Model to Use: Haiku, Sonnet, Opus or Fable

Which model should I choose? Start with a reasonable default, then compare on your own task: are facts and format correct, how much review is needed and what does the run cost? A clear brief and appropriate sources belong in the comparison too.

This is the last post in a beginner's series on Claude. The [first one](https://studiomeyer.io/en/blog/what-is-claude-2026) mapped out the whole tool. This one closes the loop with the question people obsess over most, and the more useful answer underneath it.

If you want to [bring AI into your business](https://studiomeyer.io/en/ai-solutions), we start with your tasks and ways of working before choosing the tools.

## Meet the Four: Haiku, Sonnet, Opus, Fable

Haiku, Sonnet, Opus and Fable are model families, not four unchanging model versions. Available choices depend on plan, surface and rollout. Choose for your task and check the current model list rather than treating an old ranking as fixed.

Haiku is the quick one. It is cheap and fast and perfectly good for short, simple jobs where you do not need deep thought. Sonnet is the balanced one, the everyday default, strong at almost everything including code and multi-step work. Opus is the heavyweight of the mainstream models, the one you want when a task is genuinely hard and you care more about the quality of the answer than the speed of it. And Fable, released in June, sits above all of them, the strongest model Anthropic has put in public, built for the longest and most demanding work.

That is the whole lineup. Fastest, balanced, heavyweight, strongest.

## The Part Nobody Tells You

In our experience, a clear brief and relevant context can matter more than switching models for everyday work. This is not a documented blind test or a fixed percentage of all tasks. Compare models on two or three of your own tasks using the same criteria. That shows when extra effort actually helps. If you want to keep relevant context for recurring tasks, [AI memory for your team](https://studiomeyer.io/en/services/memory) is another building block alongside your model choice.

A stronger model can matter on long or difficult work: complex reasoning, larger code changes or many dependent steps. How often this is worthwhile cannot be quantified without your task mix. Start with a balanced model and escalate when it fails clear criteria.

## The Bigger Lever

If the model matters less than you think, what matters more? How hard you let it think.

Check thinking and effort controls for the model you choose. Some current models always use thinking, so you adjust effort. Others have a toggle or API-specific options. More reasoning can improve a difficult task, but guarantees neither correct sources nor superiority over every stronger model. Test its value on your task.

## A Heuristic You Can Actually Use

Default to the balanced model for almost everything. When a task is quick and throwaway, drop to the fast one and do not think twice. When a task is genuinely hard, long, or high-stakes, reach for the heavyweight or the strongest, and turn on the deeper thinking. That is it. You do not need a decision tree. Match the weight of the model to the weight of the task, and let it think harder when the problem is actually hard.

On the free plan you do not even make this choice, you get a sensible default. Paid plans let you pick and let you think longer, which is most of what you are paying for once you are past the free tier.

## Why This Is the Wrong Thing to Obsess Over

Model capability, context, tools and a clear goal work together. A larger model cannot reliably replace missing client facts; good memory does not automatically solve a difficult reasoning task. First check material and access, then compare suitable models.

The newest, strongest model will always be one announcement away, and it will always feel like the upgrade that changes everything. It will not. The upgrade that changes everything is building the system around the model, which is exactly the stuff the rest of this series was about.

## Where to Start This Week

Stop picking the biggest model out of habit. Use the balanced one, write it a real brief, give it the context and the documents it needs, and check the appropriate reasoning effort when the problem is hard. Then compare that to whatever you were getting before from the strongest model and a one-line prompt. The gap will tell you where your leverage actually is.

That is the end of the series. If you want to go back through any of it in a structured way, our free [StudioMeyer Academy](https://studiomeyer.academy) covers the whole path lesson by lesson. The models will keep getting better on their own. The part that is up to you is everything around them.

## Sources and product status

[Claude model, thinking and effort settings](https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings) / [Fable and Mythos 5.1: current status](https://www.anthropic.com/claude-fable-and-mythos-5-1)

## Rather have it done for you?

We build exactly these systems for businesses, personally from Palma.

[See AI Solutions](https://studiomeyer.io/en/ai-solutions)

[Talk about your project](https://studiomeyer.io/en/contact)

## Read more, Claude + Claude Code

- [Can Claude Read Documents? Analyze PDFs and Contracts](https://studiomeyer.io/en/blog/analyze-documents-with-claude.md)
- [Claude Design: Prototypes, Exports and Honest Limits](https://studiomeyer.io/en/blog/claude-design-guide.md)
- [Claude Fable 5 Is Two Models Wearing One Name](https://studiomeyer.io/en/blog/claude-fable-5.md)
- [Claude in 2026: Models, Surfaces and Practical Limits](https://studiomeyer.io/en/blog/claude-guide-2026.md)
