Claude Sonnet vs Opus vs Haiku: Which to Use

Claude Sonnet vs Opus comes down to one trade: Sonnet is faster and lighter on your plan's usage limits, while Opus is stronger on work that needs judgment, according to Anthropic's Sonnet launch post and its Claude Academy model guide. Use Sonnet for everyday writing, email and well-scoped tasks, switch to Opus for proposals, strategy and analysis you will have to defend, keep Haiku for quick repetitive jobs, and save Fable, the model above Opus, for long high-stakes projects, because it uses up your limits fastest.

This guide is for people who use Claude in the apps for real work, not for developers wiring up the API. As of October 21, 2026, the current models are Haiku, Sonnet, Opus and Fable, all four listed in Anthropic's models overview, and each shipped a new version between September 1 and October 7, 2026 (dated and sourced in the box that follows). You get what each model is for, a one-table rule by job, how fast each one uses your plan, which plans include which models, and a ten-minute test that settles Sonnet vs Opus on your own work.

Claude Sonnet vs Opus: what's the actual difference?

Opus is Claude's stronger model for work that needs careful judgment, and Sonnet is the faster, cheaper model for well-scoped everyday tasks. In Anthropic's own words from the latest Sonnet launch post, Opus "is built for complex work requiring careful judgment," while Sonnet "is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets."

The quality gap is smaller than the names suggest. In the same launch post, Anthropic reports that on GDPval-AA, a test of real-world work across 44 occupations, the new Sonnet scores two points below the new Opus, and that on several evaluations Sonnet at Max effort performs comparably to Opus. Anthropic still says Opus "remains clearly stronger at complex, open-ended work requiring sustained judgment." In practice, the gap shows up less on a routine client email and more on a recommendation memo built from messy inputs, which is where Opus earns its extra cost.

Haiku and Fable sit on either side. Anthropic's latest Haiku launch post calls it "the cheapest, fastest, and most capable small model" the company has released, designed for "high-volume, cost-sensitive tasks" such as summaries and classification. Fable sits a tier above Opus: the models overview describes it as "for demanding reasoning and long-horizon agentic work" and tells developers to use it when Opus at higher effort still falls short. A fifth model, Mythos, has the same capabilities as Fable but, per Anthropic's model selection guide, is only offered to organizations verified through Anthropic's verification programs.

Context size is not a reason to choose between them in chat. On paid plans, all four current models work with a 1-million-token context window in Claude chats, according to Anthropic's context window article, and the models overview puts 1 million tokens at roughly 555,000 words on the current tokenizer.

One detail trips people up: Anthropic gives two different starting points. Its developer documentation tells anyone unsure which model to use to start with Opus "for most workloads," which is advice for teams building on the API, where the capability-first approach is to start strong and optimize cost later. Anthropic's free learning site, Claude Academy, has a model guide for app users that says "Sonnet is the daily driver" and "If you're not sure which model to pick, start here," because in the apps every Opus message spends more of your plan. For people using Claude at work, the second answer is the one that fits.

This guide is about choosing inside Claude. If you are still choosing between assistants, read Claude vs ChatGPT for real work or our three-way comparison with Gemini.

Which Claude model should I use? The one-table rule

Use the lightest Claude model that does the job well: Haiku for mechanical tasks, Sonnet for most writing and everyday work, Opus when a mediocre answer would cost you real time or credibility, and Fable only when a long, many-step job has already beaten Opus. That order follows Anthropic's Claude Academy model guide, which says to save Sonnet, Opus and Fable "for the work where the lighter answer misses things that matter."

Our shortcut for the Sonnet vs Opus call is the cost-of-a-bad-answer test. If fixing a weak answer would take you more than ten minutes, or the output goes to a client or your boss with light edits, start on Opus. If you would rewrite it anyway, or it is one step in a longer process, start on Sonnet. Here is how that plays out by job:

Email and quick replies

For triaging an inbox, summarizing a long thread or drafting a two-line reply, Haiku is enough. The Claude Academy guide recommends it for simple questions with short answers, quick lookups, pulling specific information out of text and simple summaries, and says Haiku is "the most efficient with your rate limit." Move to Sonnet when the email carries tone, such as pushing back on a client or delivering bad news, and to Opus when the message is a negotiation where every sentence matters.

Which Claude model is best for writing?

Sonnet is the best default for writing, and Opus is worth it for the pieces where structure and judgment matter most. The Claude Academy guide lists writing and content creation as Sonnet work, and the latest Sonnet announcement says it is strongest at creating polished documents, slides and spreadsheets. The current Opus, by Anthropic's description in its launch announcement, "communicates more naturally than prior models" and "puts the most important information up front," which is what a proposal, an executive summary or a strategy memo needs.

A workflow that keeps both quality and usage in check: draft on Sonnet, then switch the same conversation to Opus for the final pass. Anthropic's model and effort settings article confirms you can change the model at any point in a conversation, and the change applies from Claude's next response. A final-pass prompt for that switch is in the testing section below. For the voice setup that makes either model sound like you, see our guide to writing with Claude.

Analysis, spreadsheets and research

Use Sonnet for the mechanical parts of analysis and Opus for the conclusions. The Claude Academy guide puts analysis that needs reasoning but is not extremely complex under Sonnet, and deep research and analysis you will question, redirect and build on under Opus. Anthropic's latest Opus announcement calls it "a reliable and adept researcher" that is also "strong in financial analysis and business work." A split that works: let Sonnet clean the data, write the formulas and build the charts, then ask Opus what the numbers mean and what you should do about them. Claude Academy suggests Fable for outputs built from dense source material such as long documents, charts and technical diagrams. If your numbers live in a spreadsheet, our Claude for Excel guide covers the add-in.

Cowork tasks you hand off

For multi-step tasks you hand to Claude to work through on its own, match the model to how much judgment the task needs, because these tasks cost more than chat. Anthropic's Cowork help article says multi-step tasks, like ones that run code or create files, use more of your usage than a quick question, and that task hand-off is a paid-plan feature. Since September 16, 2026, Cowork has been merging into regular Claude chats, starting with Pro and Max, so once the change reaches your plan there is no separate mode to switch to: Claude works out what a task needs. Sonnet suits well-scoped tasks, such as turning a folder of notes into a formatted report. Opus is the better pick for long jobs that need judgment along the way, and the Claude Academy guide suggests Fable for work where you describe the outcome and let Claude plan the steps. New to task hand-off? Start with what Cowork is and how it works.

Claude Code

In Claude Code, Opus is already the default on Pro, Max, Team and Enterprise, and no plan defaults to Fable, according to Anthropic's model configuration docs. Two settings are worth knowing even if you never write code yourself: type /model sonnet to switch to Sonnet for routine changes (Claude Code also saves it as your default for new sessions), or pick opusplan, which uses Opus while planning and switches to Sonnet to carry out the plan. On some plans and seats Fable bills to usage credits, and the model picker marks it "Requires usage credits." Claude Code shares its usage limits with chat, so an Opus-heavy coding afternoon shrinks what is left for your writing. Our Claude Code pricing guide covers what each plan gets.

Sorting, tagging and extraction at volume

When the job is the same small task repeated hundreds of times, such as tagging survey responses, pulling fields out of invoices or categorizing support tickets, use Haiku. Anthropic designed the newest Haiku for "high-volume, cost-sensitive tasks," and its launch post lists summaries and classification among the quick, repetitive workloads it handles reliably. It is also the first Haiku with an adjustable effort setting, and Anthropic's effort settings article lists it among the models with an effort selector in Claude, so you can raise its effort before reaching for Sonnet.

Going the other way pays off too. Once a recurring task works, run it once on the next model down; if you cannot tell the difference, keep the cheaper model. The effort setting sits in the same menu as the model, next to the send button, as Anthropic's settings article shows, and our Claude prompting guide explains when to raise it.

Whichever model you pick, give it better instructions

The Claude Starter System is a free PDF: 20 copy-paste prompts, installable skills, and quick guides that get finished work out of Claude on any plan and any model.

Download for free

How fast does each model use up your plan?

Claude Haiku uses the least of your plan's limits, Sonnet a moderate amount, Opus a lot and Fable the most, according to Anthropic's Claude Academy model guide. The guide says Opus "uses more of your rate limit, so you want to reserve it for tasks that really need it," and warns that using Opus or Fable on a task Sonnet or Haiku could handle may use more of your limit than you need to.

Anthropic ranks the models but does not publish an exchange rate, so any claim that one Opus message equals a fixed number of Sonnet messages is a guess. What Anthropic does say: Fable models "draw from your plan's regular weekly usage limits and use them faster than other Claude models," per its article on Fable models by plan, and its usage credits article advises using "our most efficient Haiku model, or the most recent Sonnet model for most tasks," because they have lower token costs than Opus.

The model is only one of the dials. Anthropic's usage and length limits article says usage also depends on the effort level and on the length and complexity of your conversations, and adds that "tools and connectors are token-intensive." Anthropic's effort settings article is blunt about the trade: higher effort means responses that "take longer and use more tokens, so you'll reach your usage limits faster," while Low and Medium suit routine tasks and stretch your usage further. Paid plans have a session limit that resets every five hours plus a weekly limit, so heavy Opus use early in the week can leave you short later in it. For the full mechanics, read how Claude usage limits work.

There is one place where the cost difference is exact: usage credits. When you go past your plan's limits and turn on usage credits, or use Fable on Pro, you pay standard API rates, which Anthropic lists on claude.com/pricing. Per million tokens, the current Haiku costs $0.10 in and $0.50 out for prompts up to 100,000 tokens, Sonnet costs $2 and $10, Opus costs $4 and $20, and Fable costs $10 and $50. Per token, that makes Sonnet 20 times the price of Haiku, Opus twice the price of Sonnet, and Fable two and a half times the price of Opus.

Two things follow from that math. A single task is cheap on any model, so for one important document, use the best model without guilt. Volume is what adds up: run that same summary 20 times a working day for a month, about 400 runs, and it comes to roughly $1 on Haiku, $19 on Sonnet, $38 on Opus and $96 on Fable. The thinking caveat comes from Anthropic's thinking documentation, which says the tokens Claude spends reasoning are billed as output tokens, and the models overview lists thinking as always on for the current Opus and Fable. Real chats also carry earlier messages, tools and connectors, so treat these numbers as a floor.

Inside your subscription you pay in limits rather than dollars, and the usage ranking points the same way as the prices: the model you leave selected for routine work matters more than the one you pick for the big piece. If you still hit limits most weeks after moving routine work to Sonnet, compare Claude Pro vs Max before paying for credits.

Which Claude plans include Opus, Sonnet, Haiku and Fable?

Every Claude plan includes Sonnet and Haiku, Opus starts with Pro at $20 a month, and Fable runs on usage credits on Pro but is included up to half of your weekly limits on Max and on Team premium seats. That is the model lineup on Anthropic's pricing page as of October 21, 2026, and the Fable plan article fills in the details.

In plain terms, Free costs $0 and gives you Sonnet and Haiku, with no Opus or Fable. Pro costs $20 a month, or $17 a month billed annually ($200 up front), adds Opus and Claude Code, and gives you at least five times Free's usage per five-hour session, all per Anthropic's pricing page. Fable is not part of Pro's included limits; you reach it with usage credits, which let you pay for usage beyond what the plan includes.

Max costs $100 or $200 a month for five or 20 times Pro's per-session usage, according to Anthropic's Max plan article, and the Fable plan article says Max lets you spend up to 50% of your weekly limits on Fable at no extra cost; after that you can keep going on usage credits or switch to another model. On Team, per the pricing page, a Standard seat ($20 a seat per month billed annually, $25 monthly) includes Opus but reaches Fable only through usage credits, while a Premium seat ($100 annually, $125 monthly) gets five times a Standard seat's usage and the same Fable allowance as Max. Enterprise costs $20 a seat per month billed annually with a 20-seat minimum, plus usage at API rates, and Fable access depends on your seat type and how your organization is billed.

Before you judge any model, check which one you are actually using: the model name next to the send button is the one your next message goes to, and clicking it opens the picker, with More models for the rest of the list, as Anthropic's model settings article shows. For the full feature-by-feature breakdown, see our Claude pricing guide, and if you are buying for a group, our Claude Team plan guide covers Standard and Premium seats.

Want help matching the right model to your actual work?

• 1,300+ non-technical professionals learning together
• Claude, Claude Cowork & Claude Code training
• The full 550+ prompt & skill Vault included with Premium
• A weekly live call to get unstuck

Join the Academy

How to test Sonnet vs Opus on your own work

The fastest way to settle Claude Sonnet vs Opus for your own work is to run one real task on both and compare the results blind. Anthropic's Claude Academy guide suggests the same habit with Haiku and Sonnet: "Run this prompt with Haiku first, then start a new chat with Sonnet and run it again," then look at where the answers differ rather than at how long they are.

1. Pick a task you do every week, with a real input: last week's meeting notes, a client brief, a data export.
2. Run your usual prompt on Sonnet in one chat and on Opus in a fresh chat, at the same effort level.
3. Paste both outputs into a third chat with the comparison prompt below, labeled A and B so the judge does not know which model wrote which.
4. If the verdict is that a busy reader would not notice, use Sonnet for that task from now on. If Opus wins clearly, keep Opus for that task and nothing else.

I'm comparing two drafts of the same task. You don't know which
model wrote which, so judge only what's on the page.

The task was: [paste your original prompt]
A good result must: [3-5 criteria, e.g. "leads with the
recommendation", "uses only numbers from the attached data",
"under 300 words"]

Draft A:
[paste]

Draft B:
[paste]

1. For each criterion, say which draft does it better and quote
   the line that shows it.
2. List any factual claim in either draft that the input
   doesn't support.
3. Would a busy reader notice the difference between A and B?
   Answer yes or no first, then explain in two sentences.

Write the result down. After two or three weeks you will have a short list of tasks that need Opus, and everything else can run on Sonnet or Haiku. Run the test again when a new model ships: the Claude Academy guide suggests spending a few minutes running your go-to tasks across models, because "each new model brings its own strengths."

For the draft-on-Sonnet, finish-on-Opus workflow from the writing section, switch the model in the picker and send this in the same conversation:

Final pass. Keep my voice and structure, and don't add any
claims that aren't already in the draft or my notes.

1. Name the three weakest points in this draft, ranked by how
   much they would hurt the reader's decision.
2. Fix them.
3. Return the full revised draft, then a three-line change log
   so I can see what you touched.

Frequently Asked Questions

Why did Claude switch to a different model than the one I picked?

On some biology and security topics, Claude may switch to a different model than the one you picked, and it tells you when it does, according to Anthropic's Claude Academy model guide. Anthropic's Fable and Mythos announcement says certain cybersecurity and life-sciences requests are routed to Opus models. It comes from Anthropic's safeguards, not from a problem with your account or plan.

Which Claude model does voice mode use?

Voice mode runs on Opus, Sonnet and Haiku. On the Free plan you get Haiku and one connected tool, while paid plans get more models and all connected tools, and you can switch models mid-conversation from the model picker, per Anthropic's July 23, 2026 voice mode announcement. Voice mode is in beta on mobile, desktop and web.

What is Claude Mythos, and can I use it?

Mythos is the same model as Fable with more permissive safeguards for vetted cybersecurity and life-sciences work. It is available only through Anthropic's trusted access programs, the Cyber Verification Program and the Life Sciences Verification Program, and for now only to a set of US organizations, per Anthropic's Fable and Mythos announcement. On a regular paid Claude plan, the top model you can use is Fable, which Anthropic's model selection guide calls its most capable model open to all customers.

Claude Haiku vs Sonnet: is Haiku good enough for everyday work?

For short, well-defined tasks, yes, and it saves the most usage: Anthropic's Claude Academy guide says Haiku and Sonnet are more efficient with your limit because "you can ask more with fewer tokens," and Haiku is the lightest of the two. For writing and multi-step work, the same guide points to Sonnet instead. Haiku also has half the room in Cowork tasks: Anthropic's context window article lists 500K tokens for the current Haiku in Cowork against 1M for the current Sonnet.

Does Opus or Fable know more recent information than Sonnet?

No. The current Haiku, Sonnet, Opus and Fable share the same reliable knowledge cutoff, June 2026, per Anthropic's models overview, so switching to a bigger model does not get you fresher facts. For anything after that, have Claude search the web (in the new Claude experience there is no toggle; it searches when it helps), which claude.com/pricing lists on every plan, including Free.