The bigger models are slower and heavier on your limits, so picking Opus or Fable for a quick email is like taking a lorry to the corner shop. Sonnet sits in the middle, and Anthropic says its latest version performs close to the previous Opus at a lower price.
Checked on 11 September 2026 against Anthropic's own pages. The line-up changes often, so check the date before you trust a row.
Use Sonnet for most everyday jobs, switch to Opus when a wrong answer would cost you, save Fable for long and critical projects, and pick Haiku when you want a quick answer fast.
The four models, in one line each
Paraphrased from Anthropic's models page and its tutorial on choosing a model. The allowance column is Anthropic's own ranking: "Haiku is the lightest, Sonnet is moderate, Opus is heavy, and Fable uses the most."
| Model | What it is for | Speed | Allowance used |
|---|---|---|---|
| Haiku 4.5 | Anthropic's fastest model, still close to the top ones in brains. Quick answers, summaries, pulling simple facts out of text. | Fastest | Lightest |
| Sonnet 5 | The best mix of speed and intelligence, and what the tutorial calls "your daily driver". Writing, analysis, coding and multi-step work. | Fast | Moderate |
| Opus 5 | Deep research and complex reasoning that needs sustained thinking. Anthropic says it comes close to Fable 5, the version before 5.1, at half the price. | Moderate | Heavy |
| Fable 5.1 | Demanding reasoning and your largest, most critical projects, the long ones that run with fewer check-ins. | Slower | The most |
The price gap behind that last column is steep. Developers pay per token (a chunk of a word), and on Anthropic's pricing page on 11 September 2026, Fable costs $10 per million tokens you send and $50 per million it writes back, ten times Haiku's $1 and $5. In the app you pay a plan, so for you that gap shows up as allowance gone faster.
How to switch model, effort and thinking
All three settings sit in the same place, per Anthropic's help page on the picker.
- Click the model name next to the send button. That opens the picker with the current models at the top. Older ones, such as Opus 4.8 and Sonnet 4.6, sit under More models, and you can ignore them while you are finding your feet. Glance at the name every time you open a new chat, so you always know which model you are on.
- Set the effort level. It runs from Low up to Max, with High as Anthropic's recommended default, and not every model shows every level. Higher effort makes Claude work harder on each answer, and Anthropic says it uses more tokens, so you reach your limits sooner. Haiku has no effort setting, so if you cannot find it, check which model you are on.
- Decide on the thinking toggle. It is separate from effort, and it gives Claude room to reason before it answers. It also uses your allowance faster, and on Opus 5 and Fable 5.1 it cannot be turned off, which is one reason those two are heavy.
- Check Settings, then Usage, before a big job. Limits reset on a rolling five-hour window and paid plans add a weekly limit on top, per Anthropic's usage help page, so a quick look tells you whether you have room for a long Opus or Fable session.
Start on Sonnet, step up when the task earns it
- Start every new chat on Sonnet. It is the best balance of speed and brains in the picker, so it handles most jobs well and saves your heavier allowance for the day you need it.
- Drop to Haiku for quick, throwaway jobs. A one-line answer, the gist of an article, a list of dates pulled out of an email. A job that simple gains little from a slower model, so take the fastest one and keep your allowance.
- Step up to Opus when a wrong answer would cost you. Comparing two job offers, checking the small print on a big purchase, a research question with money riding on it. Ethan Mollick's July guide to picking an AI makes the same call: when the stakes are high, use the most advanced models, which in Claude means Opus and Fable, set to at least High. On the Free plan Opus is not in your picker, so Sonnet is your top rung.
- Keep Fable for long, critical projects. A business plan you build over several days, a big report stitched from many sources. It uses the most allowance and costs extra on Pro, so it has to earn its place.
- Step back down when the hard part is done. Once Opus has worked out the answer, the tidy-up edits and the follow-up emails go back to Sonnet. Anthropic's help page for its coding tool, Claude Code, gives the same advice for its bigger models: switch to one when you need it, then switch back.
Look up your job
| Your job | Start on | Because |
|---|---|---|
| A quick question, a recipe swap, a word on the tip of your tongue | Haiku | Speed is the whole point, and it is the lightest on your allowance |
| Summarise a PDF or a long article | Haiku for the gist, Sonnet if you will act on it | A gist is simple; a summary you will make decisions from needs more care |
| Rewrite an email or a message | Sonnet | Writing is Sonnet's everyday job, and an email rarely needs deep reasoning |
| Plan a trip or a week | Sonnet | Multi-step, but low-stakes, and you will tweak it as you go anyway |
| Compare two job offers or check the small print before you sign | Opus, effort High | A missed clause costs real money, and this is careful reasoning over long documents |
| A big research report from many sources | Opus | Deep research is the job Anthropic's tutorial points it at |
| A long project you build over several days | Fable on Max, Opus on Pro, Sonnet on Free | Fable is made for long, critical work, but on Pro it runs on usage credits (extra usage you pay for on top of your plan), so Opus is the sensible stop |
Where the line is
- The honest bit: the top models are closer than the price gap suggests. One developer on Hacker News moved a customer's project from Opus 5 to Sonnet 5 and wrote that they "didn't notice a decrease in quality". Another, trying Fable, said that at twice the cost it had not given "noticeably better results compared to Opus." Two voices make an anecdote, and they match the rule above: check the bigger model actually earned it. Anthropic's developer docs suggest starting on Opus 5, but that advice is written for people building apps. In the app your allowance is what runs out, which is why this guide starts on Sonnet.
- This table has a shelf life. Anthropic's models page lists Haiku 4.5 for retirement no sooner than 15 October 2026, and Sonnet 5, Opus 5 and Fable 5.1 all launched between 30 June and 1 September, so the names in your picker may have moved on by the time you read this. Treat the ladder as the lesson and the names as this month's.
Not sure which row you are? Let Claude pick
Paste this into a new chat on Sonnet, fill in the brackets, and it asks what it needs, weighs your job against the four models, and hands back one model and one effort level with a reason. It works on the Free plan too. The model list inside is dated, so if your picker shows something different, it is told to trust your picker.
You are my model picker for Claude, the reason I never spend a week's allowance on an email or trust a salary decision to the lightest model. Your one job in this chat is to match the task below to the lightest model and effort level that will do it well, and to tell me straight when a bigger model is worth it. THE MODELS YOU ARE CHOOSING FROM (checked on 11 September 2026) - Haiku 4.5: the fastest and the lightest on my allowance. Good for quick answers, summaries and pulling simple facts out of text. - Haiku has no effort setting, so if you pick Haiku, say "no effort setting" instead of a level. - Sonnet 5: the everyday model, fast and strong. Good for writing, rewriting, analysis, planning and most multi-step work. - Opus 5: slower and heavy on my allowance. For complex reasoning, deep research and anything where a wrong answer would cost me. - Fable 5.1: the slowest and the heaviest on my allowance. For long, critical projects that run over many steps with few check-ins. - On Sonnet, Opus and Fable, effort runs from Low up to Max (Anthropic recommends High as the default), and not every model shows every level. Higher effort uses more of my allowance. - The thinking toggle is separate from effort. It cannot be turned off on Opus 5 or Fable 5.1, and it also uses my allowance faster. - Plans: Free gets Sonnet and Haiku. Pro adds Opus, its strongest model, and Fable only runs on usage credits outside the plan's limits. Max 5x and Max 20x get all four, with Opus 5 as the default and Fable capped at half of the weekly limits. - Limits reset on a rolling five-hour window, and paid plans add a weekly limit on top. My real numbers live in Settings, then Usage. - Haiku 4.5 is flagged for retirement no sooner than 15 October 2026. If it is missing from my picker, use Sonnet at Low effort for the quick jobs. - Older versions, such as Opus 4.8 and Sonnet 4.6, sit under More models. Prefer the current version of the same name unless I give you a reason. - Claude Mythos 5.1 is the same model as Fable 5.1 for vetted organisations only. If I ask, say in one line that I cannot pick it. HOW TO USE THIS LIST - Your own memory of Claude's models and plans is probably older than this list, and some Anthropic pages disagree about which plan gets Fable. Where they differ, use this list. - If my picker or Settings, then Usage shows something different, trust what I see and tell me which line is out of date. - If you know today's date and it is more than two months after 11 September 2026, or I name a model missing from this list, say the list may have moved on and ask me to read out the names in my picker before you pick. Do not guess what a new model does. - Never estimate how much allowance I have left. If the pick depends on it, send me to Settings, then Usage. THE TASK I WANT TO DO [Describe the actual job in two or three sentences: what you will hand Claude and what you want back at the end. Name the real material, because "help with my CV" and "rewrite my CV for a head of marketing role using this job ad" need different models. For example: I want to turn a month of meeting notes, about 40 pages, into a one-page list of the decisions we made and who owns each one.] WHAT IS RIDING ON IT [Say what happens if the answer comes out wrong or thin: money, a relationship, a deadline, or five minutes of your time. Be honest, because this line decides most of the recommendation. For example: my manager forwards it to the client on Monday, so a missed decision would be embarrassing and hard to walk back.] HOW BIG IT IS [How much material there is and how long the job runs: one quick message, one sitting with a long document, or something you will come back to over several days. Count the pages or files if you can. For example: three PDFs of about 15 pages each, done in one sitting, with maybe two rounds of edits after.] HOW SOON I NEED IT [Say whether you need the answer in the next few minutes or can leave Claude working while you make a coffee. The lighter models answer faster, so this can tip a close call. For example: I need it before a call in twenty minutes, so a slow, careful answer is no use to me today.] MY PLAN AND WHAT I HAVE LEFT [Free, Pro or Max, and how much of your allowance you have used, from Settings, then Usage. If you have not looked, say so, and the answer will allow for it. For example: Pro, and Settings, then Usage says I have used most of this week's limit already, so I need to be careful until it resets.] WHAT I HAVE ALREADY TRIED (leave blank if nothing) [If you already ran this on a model and it fell short, say which one and how. "Sonnet missed two of the dates in the contract" is a reason to step up; "it was fine but slow" is a reason to step down. For example: I tried it on Haiku and the summary skipped the action points at the end of each meeting.] WHAT I MIGHT BRING YOU - A NEW JOB. Run the four steps below. If the job needs something Claude cannot make, such as a video or a voiceover, tell me no model here fixes that. - A JOB IN TWO HALVES. If there is a hard part and an easy part, pick a model for each and tell me when to switch. The write-up and follow-ups go back to Sonnet. - A CHAT THAT IS GOING WRONG. Ask for one example of what it got wrong. If it missed care or reasoning, step up one rung or raise effort one level, whichever costs less. If my request was vague, say so and give me the one line to add before I spend more allowance. If it was fine but slow, step down. - MY ALLOWANCE IS NEARLY GONE. Sort my jobs into what can wait for the five-hour reset, what runs on Haiku or Sonnet at Low now, and the one job that deserves what is left. - SHOULD I CHANGE PLAN. Name the one job that would change, and send me to Anthropic's pricing page for the current price. Never quote a price from memory. BEFORE YOU ANSWER, work through these four steps in order, and do not show me the working: 1. Check you have what you need. If the stakes, the size of the job or my plan is missing or vague, ask me up to three short questions and give me options to pick from, such as "If this came out wrong, would it cost you money, a relationship, or a few minutes?", "Is this one sitting, or something you will build on over several days?" and "Free, Pro or Max?". Then stop and wait for my answers. Do not recommend a model until you have them. 2. Weigh it against the ladder. Start from Sonnet at High effort as the baseline, and move only when the task gives you a reason: - If the job is short, low-stakes and speed is the point (a quick question, the gist of one article, a list pulled out of an email), step down to Haiku. - If the job is writing, rewriting or planning, Sonnet is almost always enough, even when it matters to me. - If a mistake would cost me money, a relationship or a deadline, or the job needs careful reasoning across a lot of material, step up to Opus. - Consider Fable only if the job is long, runs over many steps and is critical, and my plan allows it: Max, or Pro with usage credits I have said I am happy to spend. If it does not qualify, say in one line why Fable is not worth it here. - If I am on Free, recommend only Sonnet or Haiku. - If a smaller model already fell short, step up by exactly one rung. - If my allowance is running low, weigh that in: a lighter model now, or waiting for the reset, can beat spending what is left on one job. - For effort: Low or Medium for quick jobs, High for normal work, and the top levels only for the hardest reasoning, if my model shows them. Haiku has none. Tell me whether to switch the thinking toggle on, and remind me that higher effort and thinking both use my allowance faster. 3. Give me your pick in exactly this shape: MY PICK: the model, the effort level, and thinking on or off SWITCH TO: for a job in two halves, the model for the second half and the moment to switch (leave out otherwise) WHY: one line, tied to what I told you STEP UP IF: the one thing that would make a bigger model worth it STEP DOWN IF: the point at which I can drop back to a lighter model ALLOWANCE NOTE: one line on how heavy this is for my plan LEAST SURE: the part of this pick you are least sure of, and what I could tell you to settle it 4. Check it before you send it. Read your pick back against what I told you. If it is a close call, say so and name the runner-up. Never invent a plan limit, a price or a model detail. If the job is a legal, medical or big money decision, add one line that a better model does not replace a qualified person. If you have understood, ask your questions now, or give your pick straight away if I have already answered everything.
Here it is filled in by someone weighing up two job offers, so you can see how much to write in each part.
You are my model picker for Claude, the reason I never spend a week's allowance on an email or trust a salary decision to the lightest model. Your one job in this chat is to match the task below to the lightest model and effort level that will do it well, and to tell me straight when a bigger model is worth it. THE MODELS YOU ARE CHOOSING FROM (checked on 11 September 2026) - Haiku 4.5: the fastest and the lightest on my allowance. Good for quick answers, summaries and pulling simple facts out of text. - Haiku has no effort setting, so if you pick Haiku, say "no effort setting" instead of a level. - Sonnet 5: the everyday model, fast and strong. Good for writing, rewriting, analysis, planning and most multi-step work. - Opus 5: slower and heavy on my allowance. For complex reasoning, deep research and anything where a wrong answer would cost me. - Fable 5.1: the slowest and the heaviest on my allowance. For long, critical projects that run over many steps with few check-ins. - On Sonnet, Opus and Fable, effort runs from Low up to Max (Anthropic recommends High as the default), and not every model shows every level. Higher effort uses more of my allowance. - The thinking toggle is separate from effort. It cannot be turned off on Opus 5 or Fable 5.1, and it also uses my allowance faster. - Plans: Free gets Sonnet and Haiku. Pro adds Opus, its strongest model, and Fable only runs on usage credits outside the plan's limits. Max 5x and Max 20x get all four, with Opus 5 as the default and Fable capped at half of the weekly limits. - Limits reset on a rolling five-hour window, and paid plans add a weekly limit on top. My real numbers live in Settings, then Usage. - Haiku 4.5 is flagged for retirement no sooner than 15 October 2026. If it is missing from my picker, use Sonnet at Low effort for the quick jobs. - Older versions, such as Opus 4.8 and Sonnet 4.6, sit under More models. Prefer the current version of the same name unless I give you a reason. - Claude Mythos 5.1 is the same model as Fable 5.1 for vetted organisations only. If I ask, say in one line that I cannot pick it. HOW TO USE THIS LIST - Your own memory of Claude's models and plans is probably older than this list, and some Anthropic pages disagree about which plan gets Fable. Where they differ, use this list. - If my picker or Settings, then Usage shows something different, trust what I see and tell me which line is out of date. - If you know today's date and it is more than two months after 11 September 2026, or I name a model missing from this list, say the list may have moved on and ask me to read out the names in my picker before you pick. Do not guess what a new model does. - Never estimate how much allowance I have left. If the pick depends on it, send me to Settings, then Usage. THE TASK I WANT TO DO I have two job offers and need to answer both by Monday. One pays more but means an hour's commute each way; the other is hybrid with a better pension match. I will paste both offer letters and my notes from the two interviews, and I want a side-by-side comparison plus the questions I should ask each company before I say yes. WHAT IS RIDING ON IT A lot. This is my next two or three years and my salary, and if the comparison misses a notice period or a clawback clause in either letter, I could lose money I will not get back. HOW BIG IT IS Two offer letters of about four pages each and one page of my own notes. One sitting, then probably a couple of follow-up questions. HOW SOON I NEED IT By Sunday evening. I can leave it working for a few minutes. MY PLAN AND WHAT I HAVE LEFT Pro. Settings, then Usage says I have used about a third of this week's limit. WHAT I HAVE ALREADY TRIED (leave blank if nothing) Nothing yet, this is my first go. WHAT I MIGHT BRING YOU - A NEW JOB. Run the four steps below. If the job needs something Claude cannot make, such as a video or a voiceover, tell me no model here fixes that. - A JOB IN TWO HALVES. If there is a hard part and an easy part, pick a model for each and tell me when to switch. The write-up and follow-ups go back to Sonnet. - A CHAT THAT IS GOING WRONG. Ask for one example of what it got wrong. If it missed care or reasoning, step up one rung or raise effort one level, whichever costs less. If my request was vague, say so and give me the one line to add before I spend more allowance. If it was fine but slow, step down. - MY ALLOWANCE IS NEARLY GONE. Sort my jobs into what can wait for the five-hour reset, what runs on Haiku or Sonnet at Low now, and the one job that deserves what is left. - SHOULD I CHANGE PLAN. Name the one job that would change, and send me to Anthropic's pricing page for the current price. Never quote a price from memory. BEFORE YOU ANSWER, work through these four steps in order, and do not show me the working: 1. Check you have what you need. If the stakes, the size of the job or my plan is missing or vague, ask me up to three short questions and give me options to pick from, such as "If this came out wrong, would it cost you money, a relationship, or a few minutes?", "Is this one sitting, or something you will build on over several days?" and "Free, Pro or Max?". Then stop and wait for my answers. Do not recommend a model until you have them. 2. Weigh it against the ladder. Start from Sonnet at High effort as the baseline, and move only when the task gives you a reason: - If the job is short, low-stakes and speed is the point (a quick question, the gist of one article, a list pulled out of an email), step down to Haiku. - If the job is writing, rewriting or planning, Sonnet is almost always enough, even when it matters to me. - If a mistake would cost me money, a relationship or a deadline, or the job needs careful reasoning across a lot of material, step up to Opus. - Consider Fable only if the job is long, runs over many steps and is critical, and my plan allows it: Max, or Pro with usage credits I have said I am happy to spend. If it does not qualify, say in one line why Fable is not worth it here. - If I am on Free, recommend only Sonnet or Haiku. - If a smaller model already fell short, step up by exactly one rung. - If my allowance is running low, weigh that in: a lighter model now, or waiting for the reset, can beat spending what is left on one job. - For effort: Low or Medium for quick jobs, High for normal work, and the top levels only for the hardest reasoning, if my model shows them. Haiku has none. Tell me whether to switch the thinking toggle on, and remind me that higher effort and thinking both use my allowance faster. 3. Give me your pick in exactly this shape: MY PICK: the model, the effort level, and thinking on or off SWITCH TO: for a job in two halves, the model for the second half and the moment to switch (leave out otherwise) WHY: one line, tied to what I told you STEP UP IF: the one thing that would make a bigger model worth it STEP DOWN IF: the point at which I can drop back to a lighter model ALLOWANCE NOTE: one line on how heavy this is for my plan LEAST SURE: the part of this pick you are least sure of, and what I could tell you to settle it 4. Check it before you send it. Read your pick back against what I told you. If it is a close call, say so and name the runner-up. Never invent a plan limit, a price or a model detail. If the job is a legal, medical or big money decision, add one line that a better model does not replace a qualified person. If you have understood, ask your questions now, or give your pick straight away if I have already answered everything.
Here is the shape of answer to expect. This one is illustrative, and yours will depend on what you tell it.
MY PICK: Opus 5, effort High, thinking on (it cannot be switched off on Opus 5) SWITCH TO: Sonnet 5 at High once the comparison is done, to draft your replies to each company. WHY: A salary decision across two offer letters, where one missed clause costs you money, is the careful-reasoning job Opus is for. STEP UP IF: You add more documents and build on this over several days. On Pro, Fable runs on usage credits, so Opus stays the sensible stop. STEP DOWN IF: Your follow-ups turn into quick checks, such as a start date or the commute, which Sonnet handles. ALLOWANCE NOTE: Opus is heavy, and you said a third of your week is used, so do the comparison in one focused chat. LEAST SURE: Whether Sonnet at High would catch the same clauses. If the letters are short and plain, it may. A better model does not replace a qualified person. For any clause you are unsure of, ask one before you sign.
Once you have the pick, open a new chat, set the model, effort and thinking it named (the steps are under How to switch, above), and give that chat your real job. Keep the picker chat for picking only.
What your plan lets you pick
Prices as shown in US dollars on Anthropic's pricing page on 11 September 2026. UK buyers see local pricing where it is supported, per Anthropic's Pro plan help page.
| Plan | Price (USD) | Models you can pick |
|---|---|---|
| Free | $0 | Sonnet and Haiku |
| Pro | $17 a month billed annually, or $20 monthly | Sonnet, Haiku and Opus. Fable shows in your picker, but it runs on usage credits from the first message. |
| Max 5x | $100 a month | All four. Opus 5 is the default, and Fable can use up to half of your weekly limits. |
| Max 20x | $200 a month | All four, same terms as Max 5x |
On Pro, Opus 5 is the strongest model the plan includes. Anthropic's Fable help page has the detail, and Max users who pass half their weekly limit on Fable can buy credits or switch model. The Max plan page covers both Max tiers.
You may see Claude Mythos 5.1 in the news. It is the same model as Fable 5.1 with different safeguards, offered only to vetted cyberdefence and life-science organisations, so it will not appear in your picker (Anthropic's Mythos page).
If you hit your limit halfway through
- Do the heavy part first. Give Opus the reasoning while you have room, then switch to Sonnet in the same chat for the tidy-up. You can change model at any point, and the change applies from Claude's next response.
- Start a fresh chat when one gets long. Anthropic says longer conversations use more of your limit and suggests "starting a new conversation if you're approaching your usage limit in a longer chat". Ask Claude for a short summary first and paste it into the new chat.
- Drop the effort for routine turns. Low or Medium stretches your usage further.
- When you are out, you have three options. Wait for the reset, upgrade, or buy usage credits, per the usage help page.
One more tool worth knowing
If you are on a Mac and want to see your allowance without opening Settings, ClaudeMeter is a free, open-source menu-bar app made by one developer, Matthew Diakonov. It shows your five-hour session usage, your weekly quota and a count per model, so you can watch how much faster Opus uses your week than Sonnet does. It needs macOS 12 or later and works through a companion browser extension that reads your claude.ai account data, so only install it if you are comfortable with that. It is not made by Anthropic.
ClaudeMeter is open source (MIT): github.com/m13v/claude-meter.
Try it on your next task
Next time you open Claude, glance at the model name by the send button before you type. If it is an everyday job, leave it on Sonnet and get going; if it is one where a wrong answer would cost you, switch to Opus first. When you cannot tell, paste the Pick my model prompt with that one task, and in a couple of minutes you have a model, an effort level and the reason.
New to Claude altogether? Start with Your first hour with Claude. And before you reach for a bigger model to fix a video or voice job, check what Claude can't do, and what to use instead, because no model in the picker changes that list.
A few quick questions
Which Claude model is best for everyday use?
Sonnet. Anthropic describes Sonnet 5 as its best combination of speed and intelligence, and its own tutorial calls Sonnet the daily driver for writing, analysis and multi-step work. Step up to Opus when a wrong answer would cost you.
Can I use Claude Fable on the Pro plan?
Only with usage credits. Anthropic's help centre says Fable 5 and Fable 5.1 are not included in the Pro plan's usage limits, so they run on usage credits instead. Max plans include Fable, up to half of your weekly limits.
Do Opus and Fable use up my Claude allowance faster?
Yes. Anthropic says your usage depends on which model you pick and the effort level you set, and its tutorial ranks Haiku the lightest, Sonnet moderate, Opus heavy and Fable the heaviest. Higher effort and the thinking toggle add to it. Check Settings, then Usage, to see where you are.
What do I do when I hit my Claude usage limit?
Wait for the reset, upgrade, or buy usage credits. To stretch what you have, start a fresh chat when one gets long, lower the effort for routine turns, and switch to Sonnet once the hard part is done.