The moment you look at Claude’s model list, it feels like homework. Fable 5, Opus, Sonnet, Haiku, version numbers, price tags. Most people freeze and just use whatever is on by default.
You do not need to memorize any of it. You need one rule: match the model to the size of the job. Small job, small model. Big job, big model. That is genuinely the whole thing, and once it clicks you will never stress about the picker again.
Here is the map.
The four tiers, in plain English
The names and numbers will keep changing as new models ship. The tiers will not. Learn these four and you are set for every version that comes after.
Haiku, the fast and cheap one. Built for speed and volume. Quick answers, simple rewrites, sorting and tagging, anything you are doing a hundred times. It is not the deepest thinker, and for these jobs it does not need to be. When you want an answer in a blink and it does not have to be brilliant, this is it. Simple example: sorting a week of receipts into categories, or rewriting one text message five different ways so you can pick the best one.
Sonnet, the everyday workhorse. This is the one you will use most. Drafting, summarizing, planning, turning notes into something usable, most of your day-to-day. It hits the best balance of smart, fast, and affordable. If you are not sure which to pick, start here. It handles the large majority of real work beautifully. Simple example: turning your messy meeting notes into three clear takeaways, or writing a birthday party invite that doesn’t sound stiff.
Opus, the deep thinker. When the job is genuinely hard and the answer has to be right, you move up to Opus. Tricky analysis, careful reasoning, writing that carries weight, problems where a shallow answer would cost you. It is slower and pricier than Sonnet, and worth it exactly when the thinking is the point. Simple example: deciding whether refinancing your mortgage actually makes sense, or figuring out why a campaign flopped from a pile of messy numbers.
Fable 5, the flagship. The biggest, longest, most tangled jobs. The kind you would hand a person and not expect back for a day or a week. It runs long, multi-step projects start to finish and holds the thread the whole way. It is the premium tier, so you save it for the projects that earn it. Simple example: hand it three years of messy sales spreadsheets and ask it to find the one client segment quietly draining your margin, then have it build the board deck that explains why. Or hand it every document from a full home renovation, budget, contractor bids, permit deadlines, and let it run the entire project plan end to end. (If you want to get the most out of it, here is how to prompt Fable 5, because the big model wants a different kind of instruction.)
The one-look decision table
Same four tiers, with a real example from work and from home so you can see where your task lands.
| Tier | Reach for it when | At work | At home |
|---|---|---|---|
| Haiku (fast, cheap) | It is quick, simple, or high-volume | Tag 200 support tickets by topic | Sort a brain dump into a quick list |
| Sonnet (everyday) | It is normal daily work | Draft a campaign email, summarize a meeting | Plan the week’s dinners, write a message |
| Opus (deep thinker) | It is hard and has to be right | Analyze messy numbers into a real recommendation | Weigh a big financial decision carefully |
| Fable 5 (flagship) | It is big, long, and multi-step | Dig through three years of scattered sales data to find why one region is bleeding margin, then build the board deck | Run a full home renovation start to finish: budget, contractor bids, permits, timeline |
Notice the pattern down the rows: the jobs get bigger and the stakes get higher. You are not picking a “better” model each time. You are picking the right size.
How to actually switch
In the Claude app there is a model picker, usually near the message box or in settings. You can set it yourself, or let Claude choose for you. For most everyday work, letting it auto-pick is totally fine. You mainly reach for the picker in two moments:
- Moving up, when you want the heavy model on a job that deserves it.
- Moving down, when you want something fast and cheap and the answer does not need to shine.
If you use Claude through another tool or an agent someone built, the model may be chosen for you behind the scenes. That is fine too. The point of knowing the tiers is so you can ask for the right one when it matters.
The two signals that say “switch”
You will not always guess right on the first try, and you do not have to. Watch for these:
Move up a tier when the answer comes back thin. If you asked a real question and got a shallow, generic, or slightly-wrong answer, do not fight it with a longer prompt. Bump it to the next model up and ask again. A harder problem often just needs a stronger model, not more words from you.
Move down a tier when it is overkill. If a task is simple but the big model is slow and you are watching it churn on something a lighter one would nail instantly, drop down. You are paying in time and money for power you are not using.
That is the whole feedback loop. Start in the middle with Sonnet, move up when the answer is thin, move down when the model is overkill. You will calibrate fast.
Do this today
Look at the last three things you asked AI to do. Put each one in a tier: was it a Haiku job, a Sonnet job, an Opus job, or a genuine Fable 5 job? You will probably notice two things. One, most of your work is Sonnet-sized. Two, you have at least one big project you have been avoiding that is a real Fable 5 job.
That big one is worth handing over properly. Start with how to prompt Fable 5 so you brief it right, and if you are still deciding between Claude, ChatGPT, and Gemini in the first place, here is how I pick my tool.
Want my one-page model cheat sheet, the exact tasks I run on each tier, and the two switch signals in a format you can pin above your desk? Grab it below and I will send it over.