You’ve probably seen “Claude Code” mentioned everywhere. Is it just another ChatGPT? Is it Cursor in a terminal? This guide settles it, with current pricing, model versions, and a first task you can run today (figures verified June 2026).
TL;DR
Claude Code is Anthropic’s official command-line coding agent. It reads your whole repo, edits files, runs shell commands and tests, and makes git commits — all from the terminal, driven by Anthropic’s Claude models. Unlike ChatGPT, it acts on your machine; unlike Cursor, it lives in the terminal instead of an editor. It requires a paid Claude plan (Pro at $20/month and up) — the free Claude.ai plan does not include it.
What Claude Code is
In one sentence: Claude Code is Anthropic’s official command-line coding agent.
It runs in your terminal and can:
- Read and edit your entire project’s source code.
- Run shell commands, execute tests, and make git commits.
- Chain multi-step actions into agentic workflows (explore, plan, implement, verify).
- Run Anthropic’s Claude models only. On the Pro plan the default is Claude Sonnet 4.6; on Max, Claude Opus 4.7 is the default. Switch any time with
/model opusor/model sonnet.
A useful mental model: ChatGPT’s reasoning + Cursor’s code awareness + a real shell agent that actually executes commands.
How it differs from ChatGPT and Cursor
| Tool | Form factor | Edits your code? | Runs commands? | Underlying model |
|---|---|---|---|---|
| ChatGPT | Browser / app chat | No (copy-paste) | No | GPT-5.5 |
| Cursor | AI code editor | Yes | Partial (terminal tool) | Sonnet 4.6, Opus 4.7, GPT-5.5, Gemini 3.1 Pro |
| Claude Code | Terminal (plus Desktop app, VS Code/JetBrains extensions) | Yes | Yes (full shell) | Claude only (Sonnet 4.6 / Opus 4.7) |
The key difference: Claude Code isn’t a chatbot. It’s an agent that can read, change, and run things on your machine — then commit the result.
If you want a side-by-side decision on the editor question specifically, see Claude Code vs Cursor.
What you need to run it
As of June 2026, Claude Code requires a paid Claude plan or API access — the free Claude.ai plan does not include it (Anthropic support).
| Plan | Price (June 2026) | Claude Code usage |
|---|---|---|
| Free | $0 | Not included |
| Pro | $20/mo ($17/mo billed annually) | Included; shared usage pool, default model Sonnet 4.6 |
| Max 5x | $100/mo | Higher caps; Opus 4.7 default |
| Max 20x | $200/mo | Highest caps; can pay per token past the cap |
| API (Console) | Pay-as-you-go | Opus 4.7 $5/$25, Sonnet 4.6 $3/$15 per 1M tokens (in/out) |
Pro now bundles Claude Code and Claude Cowork into the same $20 subscription. Usage runs on a rolling 5-hour budget plus a weekly cap that resets seven days after a session starts; heavy daily users tend to land on Max, occasional users on Pro. Current numbers are on the official pricing page.
Who it’s for
- Developers who already live in the terminal.
- People moving past “AI writes me a snippet” to “AI ships the whole task.”
- Engineers who want AI to handle refactors, test writing, command running, and other grunt work — and to leave a clean git history doing it.
Non-developers are better served by Claude.ai (web chat) or Cursor (a friendlier editor UX).
Get productive in 10 minutes
1. Install
Claude Code needs macOS 13+, Windows 10 (1809+), or a recent Linux, plus 4 GB+ RAM. The native installer is the current recommendation:
# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bashPrefer npm? It works the same but needs Node.js 18 or later:
npm install -g @anthropic-ai/claude-codeThere’s also a Desktop app and VS Code / JetBrains extensions if you’d rather not start in the terminal. Verify the install with claude --version or, if anything looks off, claude doctor.
2. First run
From inside any project:
claudeIt opens a browser window to log in with your Claude account. (No paid plan, no access — see the table above.)
3. Try a low-risk first task
In the interactive prompt, paste:
This is read-only, so it’s safe. Watch how it greps the repo, opens the files it needs, and summarizes. Then graduate to a small write, like “add JSDoc comments to the functions in src/utils.js.” Don’t open with a giant refactor.
The Claude Code project setup guide walks through the rest of the first hour (including the all-important CLAUDE.md project file), and the Claude Code execution prompts library has tested prompts for the most common tasks.
What it’s genuinely good at
- Understanding unfamiliar code — point it at a file or a whole directory and ask for an explanation or a call map.
- Writing tests — generate unit tests for existing functions, then have it run them and fix what fails.
- Multi-file refactors — rename a concept across the codebase and commit it in one clean diff.
- Wiring up tooling — Firebase Hosting, GitHub Actions, lint/format configs, Dockerfiles.
- Generating real docs — produce a README or API reference from actual code, not placeholders.
The pattern that pays off: ask for a plan first. Press Shift + Tab to enter Plan mode (or run /plan), and Claude explores the repo and proposes an approach before writing a single line — which stops it from charging down the wrong path on a big task.
Safety and boundaries
Claude Code can execute commands, which means it can also break things by accident. Sensible defaults:
- Work with a clean git tree so any change is one
git restoreaway from gone. - Branch first; don’t let it operate directly on
mainormaster. - Keep approval prompts on for shell and git commands until you trust a specific workflow.
- Use Plan mode (
Shift + Tab) or Ask mode for read-only analysis when you just want a recommendation. - For risky work, isolate it in a separate git worktree or a container.
When something does go sideways — it edits the wrong file or overwrites code you wanted to keep — the Claude Code edited the wrong file and Claude Code overwrites existing changes notes cover the recovery path.
FAQ
Q: Is Claude Code just a wrapper around ChatGPT? A: No. It runs Anthropic’s Claude models (Sonnet 4.6 and Opus 4.7), never OpenAI’s. It also does things ChatGPT can’t: read your repo, run shell commands, and make commits. The closest comparison is Cursor’s agent mode, but terminal-native.
Q: Do I need to be a developer to use it? A: Effectively yes. Claude Code assumes terminal comfort and a real codebase to operate on. Non-developers will be happier on Claude.ai (web chat) or Cursor (editor-based UX).
Q: How much does Claude Code cost, and is there a free tier? A: The free Claude.ai plan does not include Claude Code. You need Pro ($20/month, or $17/month billed annually — which also bundles Claude Cowork), Max ($100 or $200/month), or API access. Usage is metered on a rolling 5-hour budget plus a weekly cap; past the cap on Max 20x you can opt into per-token API billing.
Q: How is it different from Cursor? A: Cursor is an editor with AI built in; Claude Code is a CLI agent that sits next to whatever editor you use. Cursor wins for in-flow autocomplete and inline edits; Claude Code wins for multi-file refactors, scripts, test runs, and git operations. Many developers run both.
Q: Can Claude Code commit and push on its own?
A: It can run git commit and git push if you approve those commands. Keep approval prompts on for git until you trust a specific workflow, then add it to your allowed list. With a clean tree and a feature branch, even a bad commit is trivial to undo.
Related
- Claude Code project setup
- Claude Code execution prompts
- Claude Code vs Cursor
- Claude Code edited the wrong file
- Claude Code overwrites existing changes
Tags: #Claude #Claude Code #AI coding #CLI