Claude Opus 5 shipped July 24. If you’re still running Claude Code on other cheaper models, you’re leaving performance on the table.
Here’s the complete setup, from installing Claude Code to running your first Opus 5 build.
Open your terminal and run:
npm install -g @anthropic-ai/claude-code
Already have it? Update to the latest version:
npm update -g @anthropic-ai/claude-code
Check your version:
claude --version
You need v2.1.220 to get all the latest features (background agents, notifications, Opus 5 support).
Launch Claude Code in any project folder:
claude
Once inside, switch models:
/model claude-opus-5
Or launch directly with Opus 5:
claude --model claude-opus-5
That’s it. You’re on Opus 5. It’s the same $5/$25 per million tokens as Opus 4.8.
Opus 5 is smarter than 4.8. Give it project context, and it gets even better.
Create a CLAUDE.md file in your project root:
Claude Code reads this file at the start of every session. Your prompts get shorter and your first-pass accuracy goes up. I saw a 30–40% prompt reduction after adding CLAUDE.md to my projects.
Test it on something real. Here’s a starter spec you can paste directly:
Build a task manager app.Stack: React, Vite, TypeScript, Tailwind CSS.Features:1. Add task form with title and priority dropdown (Low, Medium, High)2. Task list with checkboxes to mark complete3. Filter by priority and completion status4. Task count showing total, completed, and remaining5. Local storage persistence - tasks survive page refresh6. Clean modern design, mobile responsiveAlso create Supabase SQL and ensure the UI is impressive
This spec tests multi-feature coherence, state management, and local storage… all areas where Opus 5 outperforms 4.8.
On Opus 4.8, this spec typically needs 2 prompts (filter connection breaks on first pass). On Opus 5, it should complete in 1.
As your main build runs, start a second task in the background:
claude --background "Write unit tests for all utility functions in this project using Vitest"
Your main session keeps running. The background agent writes tests, commits them, and opens a draft PR when done. You get notified.
Two tasks running in parallel in the same terminal.
The model is better, and thinking is on by default. If safety classifiers trigger, it falls back to 4.8 gracefully instead of blocking your request entirely.
You’re now running the strongest Opus model Anthropic has shipped, at the same price you were paying for last month’s version.
I write about building solo with AI, one real project at a time. Follow me on Medium to get more content.
How to Set Up Claude Code with Opus 5 in 5 Minutes (The Free Upgrade That Doubles Your Output) was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.