Guide

Which Claude model should you use? Sonnet 5.5, Opus 5.5, Fable 5.1 and Haiku compared

Published 30 September 2026

Four wooden stools of different heights lined up in a bright studio, with a small set of brass weighing scales on the tallest one

Short answerStart with Claude Sonnet 5.5 for everyday work, move to Opus 5.5 when a task is long or hard, and use Fable 5.1 only when Opus at high effort still falls short. Haiku 4.5 is the cheapest and fastest. Mythos is not open to the public.

Anthropic now has five Claude names in the news: Sonnet, Opus, Haiku, Fable and Mythos. If you're not sure which one to pick, you're not alone. This guide compares them using the numbers from Anthropic's documentation, and ends with a simple rule for choosing.

The models side by side

Model	Best for	Price per million tokens (input / output)	Speed	Context window
Claude Sonnet 5.5	Everyday writing, coding, analysis, agents	$2 / $10	Fast	1 million
Claude Opus 5.5	Long, hard coding and business tasks	$4 / $20	Moderate	1 million
Claude Fable 5.1	The hardest reasoning and multi-hour agent work	$10 / $50	Slower	1 million
Claude Haiku 4.5	Quick, cheap, high-volume jobs	$1 / $5	Fastest	200,000
Claude Mythos 5.1	Cybersecurity and life-science research	$10 / $50	Same as Fable	1 million

Prices are Anthropic's API prices in US dollars. In the Claude app you pay a subscription instead, and which models you see depends on your plan. Mythos is limited to approved organisations, so it is on the table for comparison only. See our Claude Mythos guide.

What each one is like

Claude Sonnet 5.5: the everyday choice

Released on 28 September 2026. Anthropic calls it "the best combination of speed and intelligence." It writes output more than 30% faster than Sonnet 5 and costs up to 30% less for most work, according to Anthropic's announcement. Anthropic's model guide points to Sonnet for code generation, data analysis, content creation and agents that use tools. Start here. Our Sonnet 5.5 how-to covers the settings and prompts.

Claude Opus 5.5: when the task is long or hard

Released on 22 September 2026. Anthropic's guide says most workloads can start with Opus 5.5, and it is the pick for complex coding, large refactors, computer use and tasks that involve a lot of charts and screenshots. It thinks before every answer, and its default effort setting is medium. See our Opus 5.5 prompting guide.

Claude Fable 5.1: the strongest model anyone can use

Anthropic describes Fable 5.1 as its most capable model open to all customers. It shares Mythos's capabilities but comes with safety filters that can decline some requests. Anthropic recommends it "when your evals on Claude Opus 5 at higher effort still fall short", for example agent sessions that run for hours or deep multi-step research. It costs more than twice as much as Opus and is slower. See our Fable 5.1 prompting guide.

Claude Haiku 4.5: small, fast and cheap

Haiku 4.5 is the older, smaller model in the line-up. It has a 200,000-token window and a lower output limit, but it is the cheapest and quickest. Anthropic lists it for real-time apps, high-volume jobs and small helper tasks inside bigger systems.

A simple rule for choosing

Try Sonnet 5.5 first. For writing, summaries, studying, spreadsheets, and everyday coding, it is usually enough.
Move to Opus 5.5 if Sonnet gets stuck, loses track of a long task, or you need it to work through a big codebase.
Move to Fable 5.1 only if Opus at higher effort still isn't good enough and the value of the result covers the higher price.
Use Haiku 4.5 when you're doing thousands of small tasks and speed and cost matter most.

Before switching to a bigger model, try raising the effort setting on the one you have. Anthropic's guide says tuning effort is "often a better lever than switching models."

The effort setting in one minute

Effort controls how much a model thinks before it answers. More thinking usually means better answers, but slower ones and more expensive ones.

Model	Default effort
Sonnet 5.5	High
Opus 5.5	Medium
Fable 5.1	High

For chat and quick tasks, Anthropic suggests starting lower. For hard problems, raise it. The top levels, xhigh and max, are worth using only when you've seen a real gain, because replies get much longer.

A test you can run in five minutes

Don't rely on anyone's benchmark, including Anthropic's. Test on your own work:

prompt — 3 blanks

I'm going to give you the same task I gave another AI model. Do it as well as you can.

Task: [PASTE A REAL TASK YOU DO, e.g. summarise this report, fix this bug, write this email]
Material: [PASTE THE DOCUMENT OR CODE]
Format: [WHAT THE ANSWER SHOULD LOOK LIKE]
Rules: Use only the material above. If something is missing, say "not stated".

Run it on Sonnet 5.5, then on a larger model. If the answers are equally good, you've just saved money. If the bigger one is clearly better, you've found a task worth paying for. Our code review prompt and executive summary prompt work well as test tasks.

Where to go from here

Browse the Claude prompts page for free prompts that work on any of these models, read how ChatGPT, Claude and Gemini compare, or let the prompt builder write a structured prompt for you.