Quick answer
For most people, the right Claude model is Opus 5.5 for important work, Sonnet for everyday tasks and Haiku for quick, simple jobs. Save Fable 5.2 for your hardest problems. Fable 5.2 tops the benchmarks, but it costs twice as much as Opus 5.5 and Anthropic itself recommends Opus 5.5 as the daily driver. The quickest way to decide is to run the same real task through two models at medium effort and compare the quality, the time and how much of your allowance it used.
Key takeaways
- Default to Opus 5.5 for important writing, research and coding.
- Use Sonnet or Haiku for routine, repeatable or high-volume jobs.
- Call in Fable 5.2 for complex builds, stubborn bugs and technical problems other models cannot crack.
- Test on your own tasks, not on benchmarks alone.
- Try a lower effort level before a bigger model; it is often the cheaper fix.
Anthropic’s new Claude Fable 5.2 has just been called the world’s best AI model, which raises an obvious question: should you switch everything to it? This guide answers one practical question: which Claude model should you use for each kind of task? For the full background on Fable 5.2, its benchmarks and pricing, read our Claude Fable 5.2 review.
📘 Main article: This guide supports our Claude Fable 5.2 review, based on Matt Wolfe’s first look at the model. Read it for the benchmarks, prices and test results behind the advice here.
In this guide
- Which Claude model should most people use?
- What is the difference between Haiku, Sonnet, Opus and Fable?
- Which Claude model suits which task?
- When is Fable worth the extra cost?
- How do I test which Claude model is right for me?
- Should I change the model or the effort level?
- How do I use Fable as an advisor without paying for it all day?
- What mistakes should I avoid?
- Frequently asked questions
- The bottom line
- Related posts
Which Claude model should most people use?
Most people should start with Claude Opus 5.5 and only move up to Fable when a task genuinely needs it. That is not just our view. In its Fable 5.2 launch, Anthropic says that for most coding and knowledge tasks, users should start with Opus 5.5, its “strongest daily driver”, and switch to a different model as required.
Matt Wolfe reached the same conclusion after testing. He felt Fable 5.2 was a big leap for really hard problems, but not a huge one for everyday research or small coding projects.
What is the difference between Haiku, Sonnet, Opus and Fable?
Think of the Claude family as a ladder, from fast and light to slow and powerful. Each step up is generally more capable, slower and more expensive to run.
| Model family | Best thought of as | Typical strengths |
|---|---|---|
| Haiku | The quick assistant | Short replies, sorting, tagging, simple rewrites, high volume |
| Sonnet | The everyday all-rounder | Drafting, summaries, routine coding, customer replies |
| Opus (currently 5.5) | The senior specialist | Important writing, deeper research, larger coding work |
| Fable (currently 5.2) | The expert consultant | Very hard, open-ended technical problems, big builds, planning for other models |
Model names and version numbers change often, so check which versions appear in your Claude model menu. Free plans do not include Fable, according to Matt’s review; you need a paid plan such as Pro or Max.
Which Claude model suits which task?
Match the model to how hard and how important the task is, not to how impressive the model sounds. Use this table as a starting point:
| Task | Start with | Step up to |
|---|---|---|
| Tagging, sorting or renaming lots of items | Haiku | Sonnet |
| Replying to customer emails from a template | Sonnet | Opus 5.5 |
| Summarising a long report or call transcript | Sonnet | Opus 5.5 |
| Writing a blog post or sales page you care about | Opus 5.5 | Fable 5.2 only for a final review |
| A small website or spreadsheet fix | Sonnet or Opus 5.5 | Opus 5.5 on high |
| A full app, game or tool built from scratch | Opus 5.5 | Fable 5.2 on medium or high |
| A bug you have failed to fix several times | Opus 5.5 | Fable 5.2 |
| Hardware, engineering or scientific problems | Opus 5.5 | Fable 5.2 |
💡 Tip: If a cheaper model gets the task right two or three times in a row, keep it there. You only need to step up when the result is not good enough.
When is Fable worth the extra cost?
Fable earns its place when a mistake or a failed attempt costs you more than the extra tokens. According to Matt’s review, it costs $8 per million input tokens and $40 per million output tokens through the API, double Opus 5.5. In the Claude app it uses your allowance faster.
Good signs that Fable is worth it:
- The task has many steps and needs a clear plan.
- Other models have tried and failed.
- You need it to work first time, such as a game’s mechanics or a parts list for a real device.
- The task is technical and open-ended rather than routine.
Signs it is not worth it: the job is short, repetitive, mostly about tone, or something Opus already does well. In Matt’s game test, Fable 5.2 nailed the mechanics, but he preferred Opus 5.5 for the graphics, so “best” really does depend on the job.
Want the full picture? Our Claude Fable 5.2 review covers the benchmarks, prices, Anthropic’s examples and Matt Wolfe’s hands-on tests.
How do I test which Claude model is right for me?
Run a fair side-by-side test on three real tasks. It takes about 30 minutes.
- Pick three tasks you actually do: one easy, one medium and one hard. Use real examples, with any private details removed.
- Write one clear prompt for each and save it in a document so you can reuse it.
- Run each prompt in two models, for example Sonnet and Opus 5.5, or Opus 5.5 and Fable 5.2. Use the same effort level for both.
- Score each result out of 5 for accuracy, usefulness and how much editing it needed. Note the time taken and, if you can, the usage or cost.
- Pick the cheapest model that scores 4 or more for each task type, and write your choices down as a simple rule list.
For a more detailed cost check, see how to work out the real cost of an AI task and how to compare AI models on token cost.
⚠️ Warning: Do not judge a model on one lucky or unlucky run. Results vary, so run your hard task at least twice before deciding.
Should I change the model or the effort level?
Often the effort level, not the model, is the cheaper fix. Claude models let you choose how long they think. Matt’s review highlights two useful findings from Anthropic’s charts:
- On one coding benchmark, Fable 5.2 scored better on “extra high” than on “max”, and max cost more.
- Fable 5.2 on “high” scored about the same as Opus 5.5 on “max”, for less money.
So before jumping up a model, try one effort step higher on the model you already use. If that does not help, try the bigger model on medium. Our guide to choosing an AI reasoning effort level explains each setting.
How do I use Fable as an advisor without paying for it all day?
Let Fable do the thinking and planning, then hand the routine work to a cheaper model. Anthropic suggests using Fable as an “advisor” or “orchestrator”. You can copy the idea by hand, even without any technical set-up:
- Ask Fable for the plan only. For example: “Plan how to build this, as a numbered list of small tasks with clear success checks. Do not write the code yet.”
- Paste each small task into Sonnet or Opus to do the actual work.
- Go back to Fable only when stuck, with the error or problem and what you have tried.
- Ask Fable for a final review of the finished result against its own plan.
If you use AI agents or automation tools, the same pattern can be set up so the agent routes simple jobs automatically. See how to route simple tasks to cheaper AI models.
What mistakes should I avoid?
- Using the top model for everything. You will burn through your allowance quickly for little gain on easy tasks.
- Always choosing max effort. More thinking is not always better and costs more.
- Trusting launch benchmarks alone. They are chosen by the vendor. Your own test matters more.
- Switching all your workflows at once. Move one task at a time and keep notes.
- Forgetting launch-week offers end. Matt notes the extra Fable usage is temporary, so do not build a habit that only works during the promotion.
Frequently asked questions
Which Claude model is best?
On current benchmarks, Fable 5.2 is the most capable. But for most everyday work, Opus 5.5 is the better choice because it is strong and half the price.
Is Sonnet good enough for most business tasks?
For routine drafting, summaries and customer replies, often yes. Test it on your own tasks and step up to Opus 5.5 if the results need too much editing.
When should I use Haiku?
For quick, simple and high-volume jobs such as tagging, sorting or short rewrites, where speed and cost matter more than depth.
Do I need a paid plan for Fable 5.2?
Yes. According to Matt Wolfe’s review, Fable is not available on the free Claude plan.
Can I switch models in the middle of a conversation?
In the Claude app you can usually change the model from the menu. For important work, start a fresh chat with a clear summary so the new model has the full context.
The bottom line
You do not need the world’s best AI model for every job. Use Opus 5.5 as your default, Sonnet and Haiku for routine work, and Fable 5.2 when a task is genuinely hard. Spend 30 minutes testing on your own tasks and write down your rules. For the background, read our Claude Fable 5.2 review.
Related posts
- Claude Fable 5.2 Review: Is It Really the World’s Best AI Model?
- Low, High or Max: How to Choose an AI Reasoning Effort Level
- How to Route Simple Tasks to Cheaper AI Models Without Losing Quality

[…] sure which Claude model to pick? Our guide Which Claude model should you use? matches Haiku, Sonnet, Opus and Fable to everyday jobs in a simple […]
[…] Not sure which Claude model to use for a project like this? Read Which Claude Model Should You Use? Haiku, Sonnet, Opus or Fable. […]
[…] Anthropic also suggests changing the effort level before changing the model. We cover that in Low, High or Max: How to Choose an AI Reasoning Effort Level, and our plain-English version of the model choice is in Which Claude Model Should You Use? Haiku, Sonnet, Opus or Fable. […]