AI WORLD HQ
Before you start
You'll need an Anthropic account with one of these paid plans: Pro, Max, Team, or Enterprise. Opus isn't available on the free tier. You also need access through either the Claude web app at claude.ai or the Claude desktop app for macOS or Windows. If you're on the mobile app, the same model picker exists but the layout is slightly tighter, so I'll flag where that matters.
One more thing worth knowing before you begin. Anthropic makes Claude, an AI assistant you chat with to write, research, brainstorm, and solve problems. They offer several models (think of these as different engine sizes inside the same car). Opus is the biggest engine — better for long, multi-step jobs, slower and pricier than the smaller Sonnet and Haiku models. For a quick email, Sonnet is usually fine. For a research brief that takes twenty minutes of background reading, that's where Opus earns its keep.
Step 1 — Open a new chat and find the model selector
Start a fresh conversation by clicking New chat in the top-left corner of the Claude web app, or + New conversation in the desktop app. Along the top of the empty chat panel, you'll see a small model name button — it usually defaults to Sonnet 4.5 or whatever your last-used model was.
You'll know it worked when you see the current model name sitting just above the message box, not buried in a settings menu.
💬 The button itself reads the model name, so if it says Sonnet 4.5, you're in the right spot to change it.
Step 2 — Switch the model to Opus 4.5
Click that model name button and a dropdown appears listing the models your plan can use. Pick Opus 4.5 (the exact label may read Claude Opus 4.5). The dropdown closes, and the button at the top of the chat now shows Opus 4.5.
You'll know it worked when the model name above the message box now says Opus and the icon next to it usually changes color to mark the higher-tier pick.
💬 Menu path: click the model name → choose Opus 4.5 from the list.
Step 3 — Turn on tools and extended thinking if you need them
Below the message box, look for the tools row (a small set of icons or toggles). You'll typically see options for Web search, Code execution, File upload, and Research. For longer agent-style tasks, click Research or enable Web search so the model can fetch outside information rather than guessing. There's also an Extended thinking toggle in the same area — switch it on for harder reasoning, leave it off for faster replies.
You'll know it worked when a small dot or highlight appears on the toggles you've enabled, and the model confirms in chat that it has those tools available.
💬 Example setting combination: Web search = on, Extended thinking = on, Research = on.
Step 4 — Write a chunky, specific brief
This is the part that actually decides whether Opus feels magical or underwhelming. In the message box, write a brief that includes three things: the goal, the inputs, and what "done" looks like. Vague prompts get vague answers, no matter which model you pick.
You'll know it worked when your prompt could be handed to a human assistant and they'd know exactly what to deliver.
💬 "Compare the pricing pages of these three SaaS tools — [link 1], [link 2], [link 3]. Pull the per-seat price, the free-tier limits, and the annual discount. Return the result as a three-row table I can paste into Google Docs."
Step 5 — Let it run, then check in
Hit Enter (or click the send arrow) and let the model work. On heavier Opus tasks you'll see a status line above the message box — things like Searching the web… or Reading file…. This is normal. Don't interrupt unless the status has been stuck for several minutes.
You'll know it worked when the model either finishes with a complete answer, or stops to ask a clarifying question it genuinely needs answered before it can continue.
💬 If you want it to keep going without checking in, add to your prompt: "Work autonomously until the table is complete. Only stop to ask me something if you're missing data you can't find."
Step 6 — Save the output and, if useful, pin the chat
When Opus delivers, click the copy icon in the bottom corner of the message to grab the answer. If you want to come back to this conversation, click the pin icon at the top of the chat panel — pinned chats stay at the top of your sidebar across sessions.
You'll know it worked when you can scroll your sidebar and see the pinned conversation sitting above your regular chat history, ready to reopen tomorrow.
💬 Tip for follow-ups: start your next message with "Using the same brief as before…" to keep Opus on the same track without restating everything.
Common mistakes
- Picking Opus for a one-line question. It's slower and costs you more of your plan's usage allowance. A quick email rewrite belongs on Sonnet. Save Opus for jobs that involve several steps or outside research.
- Forgetting to enable Web search or Research. Without those toggles, Opus can only work from what's already in the chat. If you ask it to compare three live pricing pages and the tools are off, it will either refuse or invent. Turn them on in Step 3.
- Giving a vague brief like "research my industry." Opus needs a finish line. Always tell it what format you want the answer in — a table, a one-pager, a bullet list, a spreadsheet of links. If you don't, you'll get a wall of text you'll have to redo yourself.
- Micromanaging while it works. Constant follow-up messages break the agent's flow and reset its reasoning. Give the brief, let it run, then read the result.
- Assuming the mobile app has every toggle. The mobile app covers the model picker and most tools, but Code execution and some Research options are web- and desktop-only. If a tool is missing, switch to the desktop app for that session.