If you have opened Claude and seen names like Opus, Sonnet and Haiku, the obvious question is which one you should actually use. The good news is that you rarely need to overthink it. Most Australian businesses can make a confident choice in about two minutes by matching the model to the job in front of them. This guide covers each option in plain English, gives you a short decision test, and shows roughly what each one costs.
The short answer
Claude comes as a small family of models that trade capability against speed and cost. As a rule of thumb:
Pick Sonnet for most everyday work. It is the balanced default and handles the large majority of business tasks well.
Pick Opus when the task is genuinely hard: long documents, multi-step reasoning, tricky code, or work where a mistake is expensive.
Pick Haiku when you need speed and volume: quick classifications, short replies, and high-frequency automations where cost per task matters.
If you remember nothing else, start on Sonnet and only move up to Opus or down to Haiku when you have a specific reason. The current line-up is Claude Opus 4.8, Claude Sonnet 5 and Claude Haiku 4.5, built by Anthropic.
The three tiers in plain English
Opus: for the hard problems
Opus is the most capable model in the family. It is the one to reach for when a task needs careful reasoning across a lot of information: reviewing a 60-page contract, planning a complex migration, debugging code that several people have already given up on, or drafting analysis where the logic has to hold together. It is slower and costs more per task, so you use it where that extra quality earns its keep. For a law firm or an accounting practice, that might be the handful of jobs each week where a wrong answer could cost a client thousands of dollars.
Sonnet: the everyday workhorse
Sonnet is the model most teams should live in. It is fast enough for interactive work, strong enough for real drafting and coding, and priced so you can use it all day without watching the meter. Writing customer emails, summarising meetings, building first drafts of proposals, answering questions over your own documents: this is Sonnet territory. Nine times out of ten, if you are unsure, Sonnet is the right starting point.
Haiku: for speed and scale
Haiku is the quick, low-cost option. It works best when you are running the same small task thousands of times: sorting inbound emails, tagging support tickets, pulling a few fields out of a form, or powering a chatbot that needs to reply fast. You give up some depth compared with Sonnet, but for simple, well-defined jobs you often cannot tell the difference, and the savings at volume are large.
A two-minute way to choose
When you have a task in front of you, ask three questions:
How costly is a wrong answer? If a mistake is expensive or hard to catch, lean towards Opus. If it is cheap to fix or easy to check, Haiku or Sonnet is fine.
How much thinking does it need? A single clear instruction suits Haiku. Judgement across several steps or sources suits Sonnet or Opus.
How often will you run it? A one-off deep task can afford Opus. A job you run ten thousand times a month should be as light as it can be, which usually means Haiku.
Most teams end up using two models together: Sonnet for interactive work, and Haiku for the high-volume automations running quietly in the background. Opus comes in for the small number of jobs that genuinely warrant it.
What it costs in practice
Exact prices change, so treat the figures below as ballpark numbers to build intuition rather than a quote. The three tiers can differ in cost by more than tenfold, which is why matching the model to the job matters so much at volume.
Take a support team triaging 20,000 inbound emails a month. On Haiku, that classification job might cost in the order of $30 to $60 for the month. Push the same volume through Opus and you could be looking at several hundred dollars for no real gain, because sorting an email is not a task that needs the top model. Now flip it around. A single detailed review of a major contract might cost a few dollars on Opus, and trying to save 50 cents by running it on Haiku is a false economy if it misses a clause worth $40,000.
The pattern holds across most Australian SMBs. Spending around $200 a month on the right mix of models, aimed at the right tasks, routinely replaces work that used to take many hours of staff time. The saving is rarely about finding the cheapest model. It comes from using each tier where it actually pays off.
How Australian teams usually mix models
A typical setup looks like this. Sonnet runs the assistant your staff talk to during the day. Haiku sits behind the scenes handling the repetitive, high-frequency automations. Opus is kept for the weekly or monthly heavy lifting, such as detailed reports, complex code changes, or reviewing important documents before they go out the door. You are not locked into one choice; you route each job to the model that fits it.
If you handle sensitive information, the model tier does not change your obligations under the Privacy Act or your industry rules. What changes is which model does the work. Good practice is to keep a human checking anything that carries real risk, whichever tier produced it.
Choosing a Claude model is less about picking the single best one and more about matching the tool to the task. Start on Sonnet, drop to Haiku for volume, and step up to Opus for the hard, high-stakes jobs. If you would like help mapping your own workflows to the right models, Automata AI does exactly this for Australian businesses. Book a short call and we will walk through it with you.
