Opus, Sonnet, or Haiku: Which Claude Model Should Your Team Actually Use? (And What Sonnet 5 Just Changed)

TL;DR

When people say “we use Claude,” they're usually not using one thing — they're using one of several different models that all sit behind the same friendly chat box. Anthropic ships them in a family with three roles: Opus (the most powerful, for the hardest thinking), Sonnet (the everyday workhorse that balances brains, speed, and cost), and Haiku (the fastest and cheapest, for high-volume simple tasks). On June 30, 2026, Anthropic released Claude Sonnet 5 — a new version of the middle model that now performs close to the top-tier Opus on a lot of real knowledge work, while staying much cheaper and faster. It quietly became the new default for most people. If your team has ever wondered “are we even using the good one?” or “why is Claude sometimes brilliant and sometimes just okay?” the answer is almost always which model you're on. Here's the whole family explained in plain language, what Sonnet 5 changed, and how to make sure your team is using the right one for each job.

The Thing Nobody Tells You: “Claude” Is Not One AI

Claude is a family of models, not a single product — and which one you're talking to changes the quality, speed, and cost of every answer you get.

This is one of the most common sources of confusion for non-technical teams, and it's completely understandable. You open Claude, you type, you get an answer. Nothing on the screen shouts “you are currently using the mid-tier model.” So people naturally assume Claude is one fixed thing, and that when it does something impressive or something disappointing, that's just “what Claude is like.”

It isn't. Behind that single chat box, Anthropic runs several different models, and they genuinely differ in how smart, how fast, and how expensive they are. Picking the right one is a bit like choosing between a sports car, a family estate, and a delivery van. They're all vehicles, they all get you there — but you'd be annoyed using a delivery van for a track day, and you'd be wasting a fortune using a sports car to move furniture. The models work the same way. Once your team understands the family, a lot of “why was the AI weirdly bad at that?” moments suddenly make sense.

Meet the Family: Opus, Sonnet, Haiku

Anthropic names its models in three tiers, and the names are consistent: Opus is the biggest, Sonnet is the middle, Haiku is the smallest.

The naming comes from writing — an opus is a grand work, a sonnet is a structured medium-length poem, a haiku is short and quick. That's the whole idea, and it's a genuinely useful mental model.

Claude Opus — the heavyweight. This is the most capable, most careful model. It's the one you want for genuinely hard thinking: complex analysis, tricky reasoning, high-stakes documents, anything where getting it exactly right matters more than getting it fast. It's also the slowest and most expensive to run, because all that extra capability takes more computing power.
Claude Sonnet — the workhorse. This is the balanced middle model, and it's what most people should use most of the time. It's very capable, noticeably faster than Opus, and much cheaper — the “daily driver” that handles the overwhelming majority of real work well.
Claude Haiku — the sprinter. This is the smallest, fastest, cheapest model. It's not built to write your board memo; it's built to do simple, repetitive things at enormous volume — sorting thousands of messages, tagging incoming emails, pulling one field out of a form — instantly and for almost nothing.

Each tier also has a version number — you'll see things like Opus 4.8, Sonnet 5, Haiku 4.5. The number just tells you how recent that particular model is, the same way software has version numbers. Higher usually means newer and better. The tier name (Opus / Sonnet / Haiku) tells you the size and role; the number tells you the generation.

What Actually Changed on June 30: Claude Sonnet 5

Anthropic released Claude Sonnet 5, a new version of the middle model that closes much of the gap to the top-tier Opus — while staying far cheaper and faster.

Here's the plain-English version of why this matters. For a long time, the trade-off was clear: if you wanted the very best answers, you paid for Opus and accepted that it was slower and pricier. If you wanted speed and value, you used Sonnet and accepted that it was a step down in raw capability. That trade-off is what made people nervous about “are we using the good one?”

Sonnet 5 narrows that gap dramatically. On a lot of everyday knowledge work — drafting, summarizing, analyzing, answering questions across documents — it now performs close to Opus, and on some tasks even edges past the older top model. But it keeps the middle-tier price and speed. In other words, the “good enough for almost everything” model just got a lot closer to “as good as the best,” without the cost or the wait.

For most teams, the practical consequence is simple and reassuring: the default model you're probably already using got meaningfully smarter, and you didn't have to do anything. Sonnet 5 quietly became the new standard workhorse. If you were worried you were missing out by not paying for the premium model, that worry just got a lot smaller.

Why This Matters Even If You Never Touch a Setting

The model you're on silently sets the ceiling on what Claude can do for you — so knowing the family is knowing which lever to pull when the AI underperforms.

Imagine two people at the same company both ask Claude to “analyze this 40-page contract and flag the risky clauses.” One gets a sharp, careful answer. The other gets something shallower that misses a subtle problem. They conclude Claude is “inconsistent” or “unreliable.” But often the real explanation is that one of them was on a more powerful model than the other — because of their plan, their settings, or a default they never noticed.

This is why the family matters even to people who never open a settings menu. When Claude does something disappointing on a genuinely hard task, the first question shouldn't be “is AI just not good enough for this?” It should be “was I using the right model for this job?” A huge amount of the disappointment non-technical teams feel with AI is really a mismatch: using a fast, cheap model for a task that needed the heavyweight, or waiting on the heavyweight for something the sprinter could have done instantly.

How to Choose: A Simple Rule of Thumb

Match the model to the stakes and the volume of the task, not to a vague sense of “always use the best.”

You don't need to memorize benchmarks. Here's a rule that covers almost every real situation:

Reach for Opus (the heavyweight) when the thinking is hard and the stakes are high. A complex strategy analysis, a legal or financial document where a missed detail is costly, a knotty problem that needs careful step-by-step reasoning, an important external-facing piece you can't afford to get wrong. When quality matters more than speed or cost, use the best.
Use Sonnet (the workhorse) for basically everything else — which is most of your work. Drafting emails and documents, summarizing meetings and threads, answering questions across your files, first drafts, research, brainstorming, day-to-day help. With Sonnet 5, this covers an even wider range than before. If you're unsure, this is the right default.
Think about Haiku (the sprinter) when you're doing the same simple thing thousands of times. This tier is mostly relevant when a tool or automation is running in the background — sorting a flood of support tickets, tagging incoming leads, extracting one number from each of ten thousand invoices. You'll rarely pick Haiku by hand in a normal chat; it shows up inside automations where speed and volume matter more than depth.

The one-line summary to give your team: default to Sonnet, escalate to Opus when it's genuinely hard or high-stakes, and let Haiku live inside your automations.

What This Looks Like in Real Work

Concrete examples of matching the model to the job.

A consulting firm preparing a client deliverable. The team drafts, summarizes, and iterates all week on Sonnet — fast and cheap enough to use constantly. But for the final strategic analysis that goes in front of the client's board, they switch to Opus and ask it to pressure-test the reasoning. Cheap horsepower for the volume, premium horsepower for the moment that counts.
A support team triaging incoming tickets. Behind the scenes, a Haiku-powered automation reads every incoming message and sorts it by topic and urgency instantly, for a fraction of a cent each. Then a human, helped by Sonnet, writes the actual replies to the ones that need a thoughtful answer. Right tool at each stage.
A finance lead reviewing a complex agreement. This is exactly the “hard and high-stakes” case — so they use Opus, because a missed clause could cost real money. They'd never run this one on the cheapest model just to save a few cents.
A marketing team producing weekly content. Ninety percent of the work — drafts, variations, social copy, summaries — runs happily on Sonnet. It's fast enough to keep up with a real content calendar and good enough that the drafts need light editing, not rescuing.

The through-line: nobody uses one model for everything. The teams getting the most from Claude have quietly learned to reach for the right tier, the same way you'd instinctively grab the right tool from a drawer.

Common Misunderstandings to Clear Up

A few beliefs that quietly cost teams quality or money.

“We should always use the most powerful model.” Tempting, but wasteful. The heavyweight is slower and much more expensive, and for the majority of tasks you won't notice a difference in quality — you'll just wait longer and pay more. Save it for the tasks that genuinely need it.
“The cheap model is the ‘bad’ one, so we should avoid it.” Not at all. Haiku isn't a worse version of the same thing — it's a specialist built for speed and volume. For its job (lots of simple tasks, fast), it's the right choice, and the heavyweight would be the wrong one.
“A higher version number always means I should switch.” Usually newer is better within a tier, but you don't need to chase every release. The bigger decision is almost always which tier (Opus vs Sonnet vs Haiku), not which exact version.
“I have no idea which one I'm using, so I can't control this.” You can. In most Claude interfaces there's a model picker — often a small dropdown near the message box. And if you're ever unsure, you can literally ask Claude “which model are you?” and it'll tell you.

How This Connects to Everything Else in Claude

The model is the engine; features like Projects, Skills, and agents are the body built around it.

It helps to separate two things. The model (Opus, Sonnet, Haiku) is the raw intelligence — how well Claude thinks. The features you've heard about — Projects that hold your standing documents, Skills that teach Claude your methods, agents that run tasks on a schedule — are ways of packaging and directing that intelligence. They mostly work across whichever model you choose.

That means the two questions are separate and both worth asking: “What am I asking Claude to do, and how do I set it up?” (Projects, Skills, agents) and “How much raw thinking power does this task need?” (the model tier). A well-organized Project running on the wrong model can still disappoint; the right model with no context can still be generic. Getting both right is where the real leverage is — and choosing the model is the simpler of the two to master.

What Your Team Should Do This Week

Three quick moves to stop leaving quality and money on the table.

1. Find your model picker and look at what you're defaulting to

Open Claude and find the small dropdown that names the model — usually near where you type. Just knowing what your team defaults to is worth a lot. For most people it should be a current Sonnet (now Sonnet 5), which is the right daily driver. If you've been stuck on an older or smaller model without realizing it, this two-minute check is the highest-value thing in this article.

2. Re-run one “Claude wasn't good enough” task on the heavyweight

Think of a genuinely hard task where Claude disappointed you — a thorny analysis, a complex document. Run it again on Opus and compare. Often the “AI can't do this” conclusion was really “the wrong model was doing this.” This recalibrates your team's sense of what's actually possible.

3. Give your team one sentence to remember

Adopt a shared rule so nobody has to overthink it: “Default to Sonnet. Switch to Opus when it's hard or high-stakes. Haiku lives in our automations.” That single sentence is enough for 95% of decisions, and it turns “which model?” from a mystery into a reflex.

FAQ

Is “Claude” one AI or many?

It's a family. The name “Claude” covers several models — Opus, Sonnet, and Haiku — that share the same chat experience but differ in how powerful, fast, and expensive they are. Which one you're using changes the quality, speed, and cost of every answer.

What's the difference between Opus, Sonnet, and Haiku?

Opus is the most powerful and careful (best for hard, high-stakes work, but slower and pricier). Sonnet is the balanced workhorse most people should use for everyday work. Haiku is the fastest and cheapest, built for simple tasks done at high volume, usually inside automations.

What actually changed with Claude Sonnet 5?

Sonnet 5, released June 30, 2026, is a new version of the middle model that performs close to the top-tier Opus on a lot of real knowledge work — while keeping the middle-tier speed and cost. In practice, the default model most people use got meaningfully smarter, so the gap between “the affordable one” and “the best one” narrowed a lot.

Which model should my team use by default?

Sonnet, in its current version. It handles the overwhelming majority of real work well, quickly, and affordably. Reach for Opus only when a task is genuinely hard or high-stakes, and let Haiku power your background automations. “Default to Sonnet, escalate to Opus, automate with Haiku” covers almost every case.

What do the version numbers (like 5 or 4.8) mean?

They just tell you how recent a model is, like a software version. The tier name (Opus / Sonnet / Haiku) tells you the model's size and role; the number tells you its generation. Higher is generally newer and better, but the bigger decision is which tier, not which exact number.

How do I know which model I'm currently using?

Most Claude interfaces show a model picker — a small dropdown, usually near the message box. If you can't find it, you can simply ask Claude “which model are you?” and it'll tell you.

Do features like Projects and agents depend on the model?

They mostly work across whichever model you choose. The model is the raw intelligence; Projects, Skills, and agents are ways of directing it. For best results, get both right — the correct setup and a model powerful enough for the task.

Want help setting a simple, company-wide rule for which Claude model your team uses for which job — and making sure nobody's quietly stuck on the wrong one? The Deployed Kickstart gets your team hands-on with Claude in a single day, mapped to your real workflows. The Partner program gives you ongoing support to roll it out across the business.