GUIDES
Stop using the wrong Claude model. 📚
Sonnet, Opus, Extended Thinking — they're built for different things. Using the wrong one wastes your usage limit and gives you worse results. Here's exactly.
@thesomyanayak — Claude at Work
Stop using thewrong Claudemodel.
Sonnet, Opus, Extended Thinking — they're built for different things. Using the wrong one wastes your usage limit and gives you worse results. Here's exactly when to use each one.
Why this matters more than most people think
Claude Pro gives you a usage limit. Every message costs against it — and Opus costs more than Sonnet, which costs more than using Extended Thinking on a simple task. Most people don't know this and end up either burning through their limit too fast, or defaulting to the same model for everything and leaving quality on the table for complex tasks.
The right model for the right task isn't just about quality — it's about getting more out of your subscription. This guide breaks down exactly how to use each model so you stop wasting usage on the wrong one.
The models — what each one actually is
Sonnet is the model you should have open and running all day. It handles the overwhelming majority of real work tasks at high quality with fast response times. This is not a stripped-down model — Sonnet 4.6 scores within 1.2 percentage points of Opus on coding benchmarks. For anything that isn't genuinely complex, it's the right choice.
The practical rule: start with Sonnet on everything. Only switch when you hit a wall — when the output isn't deep enough, the reasoning feels shallow, or the task genuinely needs sustained multi-step analysis.
- Writing, editing, and rewriting — emails, reports, proposals, scripts
- Summarizing documents, meeting notes, research
- Brainstorming and ideation — first-pass ideas, outlines, angles
- Drafting communication in your voice
- Research synthesis and explaining complex topics
- Creating templates, checklists, frameworks
- Day-to-day Q&A — quick lookups, explanations, advice
- Building and iterating on Claude Skills and Projects
Lightest of the two main models. Use it freely for daily work.
Fastest model. Responses come back significantly quicker than Opus.
Opus is Anthropic's most powerful model. It goes slower and costs more of your usage limit — but on the tasks it's built for, the quality difference is real. The key is that Opus shines specifically on tasks where reasoning depth, nuance, and sustained analysis matter. On routine tasks, it's overkill and can actually be worse — it over-explains and adds complexity you didn't ask for.
The practical rule: switch to Opus when Sonnet isn't cutting it. When the stakes are higher, the problem is genuinely complex, or you need Claude to think something all the way through without shortcuts — that's your signal.
- Complex strategy work — business decisions, positioning, big planning documents
- Deep analysis where you need thoroughness, not just a good answer
- Difficult writing that requires sustained voice, nuance, and judgment
- Financial, legal, or technical analysis where precision matters
- Diagnosing complex problems — anything where the root cause isn't obvious
- Long documents where context needs to be held across thousands of words
- High-stakes decisions you need to stress-test from multiple angles
Noticeably higher than Sonnet. Reserve it for tasks that genuinely need it.
1M token context — can handle massive documents, full codebases, long sessions.
The mode people overlook
Extended Thinking —what it actually does
Extended Thinking isn't a separate model — it's a mode you can activate. When you turn it on, Claude works through the problem step by step before it responds, rather than answering immediately. You can see the thinking process unfolding in real time. The output is fundamentally different — not just more words, but more thorough reasoning that catches things a fast answer misses. It uses more of your usage limit, so save it for the right moments.
You've been going back and forth on a decision. The problem has multiple competing factors. You need Claude to find something you might have missed, not just confirm what you already think.
For complex strategic or analytical questions, seeing how Claude thinks through it is often as valuable as the conclusion. You can spot where the reasoning breaks down and push back.
Writing an email, summarizing a document, getting a quick answer. Extended Thinking slows everything down and uses more of your limit for tasks where it adds nothing.
Either use the "think" toggle in the interface when it appears, or add "Think through this carefully before responding" to your prompt. Both work.
"
Opus on a simple email is like driving a Formula 1 car to get groceries. Technically fine. Completely unnecessary. And it costs you.
The quick-reference cheat sheet — by task type
The most commonmodel mistakes.
The most common one. Opus costs more and for routine tasks it's actually worse — it over-explains, adds unnecessary complexity, and slows you down. Save it for tasks that genuinely need deep reasoning. Your default should always be Sonnet.
Extended Thinking is a mode you toggle on — not a setting to leave running. If you forget to turn it off after a complex session, you'll burn through your usage limit on tasks that don't need it at all. Treat it like a tool you pick up and put down deliberately.
Before upgrading to Opus, try giving Sonnet more context first. A lot of shallow output comes from a shallow brief, not a weak model. Add more context, be more specific about what you want, and see if Sonnet gets there. Often it does — and you keep your usage.
The flip side — people who know it costs more never use it. If you're going back and forth on a real decision, spending 10 minutes with Extended Thinking on Opus is absolutely worth it. The usage cost is real, but so is the quality difference on genuinely hard problems.
If you're several messages into a Sonnet conversation and realize you need deeper analysis, starting a new Opus conversation with the full context is usually better than switching mid-thread. Opus gets more value from a clean, well-structured brief than from picking up a long messy thread.
A prompt is only as strong as its context architecture. Always define: 1) Persona & Authority, 2) Precise Business Objective, 3) Constraints (what NOT to do), and 4) Structured Output Schema (tables, JSON, or step-by-step frameworks). Save your highest-performing prompt chains as reusable team SOPs.