Claude now has four models, and the price gap between them is wild. One costs $0.10 per million tokens. Another charges $50 for the same amount. That’s a 500x difference, and picking wrong burns real money.
If you’ve opened Claude lately and felt stuck on the model picker, you’re not alone. Anthropic’s lineup changed a lot in 2026. Fable, Opus, Sonnet, and Haiku all look similar in a dropdown. They’re not.
Here’s the fix. Each model has a clear job, and once you know which Claude model to use for what, the choice gets boring in the best way. This guide breaks down all four with real prices, then gives you a simple rule for every situation.
The Claude lineup at a glance
Before the deep dive, here’s the whole family in one table. Prices are per million tokens (a token is roughly three-quarters of a word), which is how API usage gets billed.
| Model | Best for | Input price | Output price | Context window |
|---|---|---|---|---|
| Claude Haiku 5.5 | Fast, cheap, repeatable tasks | $0.10 / M | $0.50 / M | Large enough for daily jobs |
| Claude Sonnet 5.5 | Everyday chat, drafting, analysis | $2 / M | $10 / M | 1M tokens, 128K max output |
| Claude Opus 5.5 | Serious work and coding | $4 / M | $20 / M | 1M tokens, 128K max output |
| Claude Fable 5 | Frontier coding, long-running tasks | $10 / M | $50 / M | 1M tokens, 128K max output |
One quick note before we go model by model. On the Claude app (claude.ai), you don’t pay per token. You pay a subscription and hit usage limits. The prices above matter for API users and for anyone building automations, and they also explain why Anthropic pushes certain models toward free and Pro users in the app. We’ve covered the app plans before, so here we’ll focus on what each model is actually good at.
Claude Haiku 5.5: the budget workhorse
Haiku 5.5 launched in October 2026 and replaced Haiku 4. It’s the cheapest model in the family, and it’s faster than anything else Anthropic ships.
Don’t let the price fool you. In Zapier’s testing, Haiku only failed the single most advanced task in their benchmark, and the others barely beat it. For everyday jobs, it’s genuinely good. It’s also a reasoning model now, with an effort setting you can tune up or down depending on how hard the job is.
Where it shines:
- Tagging and classifying messages at scale
- Routing support tickets to the right team
- Powering chatbots that need instant replies
- Any automation that runs thousands of times a day
Here’s the thing though. Haiku makes sense mostly through the API. In the Claude app, the savings don’t really reach you, so there’s little reason to pick it for normal chatting.
If you run automations, this is your moneymaker. At $0.50 per million output tokens, you can classify 100,000 support emails for less than the cost of a sandwich. We compared it directly in our Claude Haiku 5.5 walkthrough if you want the full breakdown.
Claude Sonnet 5.5: the everyday default
Ask people who use Claude daily, and most will tell you Sonnet is where they live. At $2 per million input tokens and $10 per million output, it’s the middle child that quietly does everything.
Sonnet 5.5 handles chat, drafting, summarizing, and analysis without breaking a sweat. It has a 1M token context window, which means you can feed it a small book and it won’t forget the beginning. Knowledge cutoff is June 2026, so it’s current on recent events too.
It’s also the most advanced model available on Claude’s Free plan. That matters. If you’re not paying for Claude at all, Sonnet is what you’re getting, and it’s a strong deal for zero dollars.
In testing, Sonnet correctly categorized a batch of customer support messages but missed two ambiguous cases that Opus caught. That’s the pattern you’ll see. It nails 95% of daily work. The last 5% is where the bigger models earn their price.
For writing emails, brainstorming, planning trips, summarizing documents, and every other “help me think” task, start here. Honestly, if you’re unsure which Claude model fits your day, start here and stay a while.
Claude Opus 5.5: the professional’s pick
Opus is the model the Zapier reviewer defaults to 95% of the time, and after looking at the specs, that tracks. At $4 per million input and $20 per million output, it costs 2x Sonnet and delivers more than 2x the reliability on hard tasks.
This is the model for real work. Coding projects. Long analysis. Anything where a mistake costs you an hour of cleanup. Opus 5.5 is also cheaper than Opus 5 was, and it uses tokens more efficiently, so the real-world cost gap is smaller than the list prices suggest.
What changed with 5.5? Anthropic closed the gap with Fable on benchmarks while keeping the price sane. For most professional coding and complex work, it matches the flagship where it counts, and for anyone deciding which Claude model deserves their work budget, that settled it.
The tradeoff is speed. Opus thinks longer before answering. For quick questions, that patience is wasted. For a refactoring job or a gnarly data analysis, it’s exactly what you want.
If you’re paying for a Pro plan and you code, Opus is probably your daily driver. Our Sonnet 5.5 review covers why some users still prefer the cheaper option, and our Opus 5.5 prompting rules help you get your money’s worth once you switch.
Claude Fable 5: the specialist
Fable is Anthropic’s frontier model, and it behaves like one. $10 per million input tokens. $50 per million output tokens. It thinks hard about everything, whether you asked it to or not.
This is the model for frontier coding: massive codebases, architecture decisions, agents that run for hours. It has the same 1M token context window as the others, plus 128K max output, so it can hold an entire project in its head.
Two quirks worth knowing. First, Fable sometimes marches to the beat of its own drum, taking creative liberties when you wanted something literal. Second, when a request trips its safety guardrails, it falls back to Opus automatically. That keeps responses safe, but it means you’re sometimes paying Fable prices for Opus-level answers.
Where Fable is genuinely useful is planning code and long-running tasks that would stump the smaller models. Zapier’s testing found Fable especially strong at exactly that.
But unless you know you need it, you probably don’t. The reviewer’s own words: if you don’t have a specific reason or an unlimited budget, Opus gets you Fable’s advantages without the astronomical usage costs. We took a long look at it in our Fable 5 review.
Which Claude model should you actually pick
Enough specs. If you’re still torn on which Claude model to open right now, here’s the decision rule, situation by situation.
Picking for everyday chat and writing
Use Sonnet 5.5. On the free plan it’s what you get anyway, and on paid plans it saves your usage limits for when you need them. It drafts, summarizes, plans, and explains better than the price suggests.
Picking for coding and professional work
Use Opus 5.5. It’s the best balance of price and performance in the lineup, and it’s what experienced users reach for by default. Fable only enters the picture for frontier-scale projects, and even then, test whether Opus handles yours first.
Picking for automations and high-volume jobs
Use Haiku 5.5, no contest. Routing, tagging, classification, chatbots: anything that runs constantly belongs on the cheapest model. You’ll save 20-100x versus the bigger options, and quality is fine for structured tasks.
One more rule of thumb
Start cheap, escalate when it fails. Build your workflow on Haiku, bump failures to Sonnet, and save Opus and Fable for the jobs that prove they need muscle. Most people do this backwards and pay for it daily.
The takeaway
The Claude lineup isn’t really four competitors. It’s a ladder, and knowing which Claude model sits on each rung is the whole game: Sonnet for daily driving, Opus for serious work, Haiku for anything that runs at scale, and Fable for the rare jobs that truly need the frontier. Check the official Claude page for current prices, then pick by job, not by hype. Your wallet will notice the difference within a week.
