Choosing a Claude Model: Haiku, Sonnet, Opus or Fable
5 min read
Open the model picker in Claude and you meet four names that sound like a poetry anthology: Haiku, Sonnet, Opus and Fable. Most people either leave the default alone forever or switch to whatever sounds most powerful and then wonder why answers feel slow. Neither is a decision. It is a habit.
This article explains what Anthropic says each model is for, and gives you a way to choose that you can apply in thirty seconds. One caution first: the model page this article draws on is Anthropic's developer documentation. The picker in your own plan may show only some of these models, and the names and defaults can change, so treat the picker in front of you as the final word on what is available to you.
The current lineup in plain words
Anthropic's models overview describes four current models, from fastest to deepest.
| Model | What Anthropic positions it for | Speed relative to the others |
|---|---|---|
| Claude Haiku 5.5 | High-volume, latency-sensitive tasks such as classification, extraction and routing | Fastest |
| Claude Sonnet 5.5 | The best combination of speed and intelligence | Fast |
| Claude Opus 5.5 | Long-running agentic coding and knowledge work | Moderate |
| Claude Fable 5.1 | Demanding reasoning and long-horizon agentic work | Slower |
Two facts apply to all four and remove some worries. They all have a 1 million token context window, which Anthropic puts at roughly 555 thousand words, and they share the same reliable knowledge cutoff of June 2026. So "which model can hold my long document" and "which one knows recent things" are not the deciding questions. What differs is speed and how much hard reasoning each is built to sustain.
For most workloads, Anthropic's guidance for developers is to start with Opus 5.5, and to move to Fable 5.1 for demanding reasoning and long-running agentic work, or when your own tests on Opus 5.5 at higher effort still fall short.
Match the model to the shape of the task
Instead of ranking models, look at your task.
Reach for the faster models when
- The task is small and repeated: sorting, tagging, extracting fields
- You will read and edit the output yourself anyway
- You are iterating quickly and want many rounds
- A wrong answer is cheap and easy to spot
Reach for the deeper models when
- The task has many steps that depend on each other
- You are analyzing a long document for what is missing or inconsistent
- You want Claude to work through a large piece of work with less supervision
- A subtle error would be expensive and hard to see
A thirty-second decision
- 1
How long is the work?
A single message and a single answer points to the faster end. A job that takes Claude many steps, or many files, points to the deeper end.
- 2
What does a mistake cost?
If you would catch it in seconds, go fast. If it could slip into a report, a contract summary or a financial model unnoticed, go deep.
- 3
Are you waiting on the answer?
If you are in a live back-and-forth, speed protects your attention. If you can start the task and come back to it, wait for the deeper model.
If two of the three point the same way, you have your answer. If they split, start in the middle with Sonnet or Opus and adjust after the first result.
Effort and thinking are a second dial
Choosing a model is only half of how much work Claude puts in. The other half is effort and thinking, which Anthropic treats separately. A fast model at higher effort and a deeper model at lower effort can land in similar places for some tasks, and they cost you different things in speed and usage. Claude's Thinking Mode: When a Slower Answer Is Worth It explains how to use that second dial.
The mistake that wastes the most time
The common error is using the deepest model for everything because it feels safer. It is not safer for tasks like reformatting a table or tightening an email, where it is slower and uses your allowance faster for no gain. The opposite error exists too: asking the fastest model to reason through a problem it was not built to sustain, then blaming the whole tool for a weak answer.
Tip
Keep a note of one task where each model felt right for you: one quick edit, one long analysis. After a week you will have your own rule of thumb, which is worth more than any comparison table, including this one.
What doesn't change with the model
A few things stay the same whichever model you pick, so do not expect a bigger model to fix them.
- A vague request still produces a vague answer. The Prompt Framework That Fixes a Vague Claude Answer applies to every model.
- Facts that changed after the cutoff are still unknown unless Claude searches. See When Claude Is Searching the Web and When It Isn't.
- You still need to check numbers, quotes and anything you will act on.
Models change faster than guides
Anthropic lists legacy models that are still available alongside the current ones, and publishes retirement commitments for each. Model names, positioning and availability move on a schedule you do not control, which is why this article describes how to choose, not which model is best. When the lineup changes, the three questions above keep working.
If you are comparing across tools, the same exercise applies to Which AI Assistant for Which Job: A Decision Table.
Official sources
Checked on October 24, 2026. Features, plans and names change often, so the vendor's own pages are the final word.