Claude Code · September 2026
Claude vs Claude Code
Same company, same models, two different products. Here is what each one is, where it runs, what it can touch, and how to tell which one your question is really about.
Two products, one name.
Claude is the assistant. It is the thing you talk to at claude.ai, in the desktop apps for macOS and Windows, in the iOS and Android apps, and through the Chrome extension. Underneath it is a family of models — Opus, Sonnet, Haiku, and the Fable line — and the same models are sold separately through the API to people building their own software. When someone says “ask Claude”, they mean this: a chat window, a file you upload, an answer that comes back.
Claude Code is a different product built on the same models. Anthropic’s own definition is the useful one: an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools. It lives in a terminal. You point it at a project directory, describe a change, and it goes and makes it — reading files, running your tests, committing when you ask. It is an agent with hands, where Claude the app is an assistant with a text box.
The confusion is fair, because the two share a login, a subscription, and a name, and because Claude Code now also has a desktop app, a web version at claude.ai/code, and VS Code and JetBrains extensions. The distinction that survives all of that: Claude answers you; Claude Code acts on a machine.
Side by side.
| Question | Claude | Claude Code |
|---|---|---|
| What it is | A chat assistant: apps and a web interface over the Claude models | An agentic coding tool: a CLI that plans and carries out changes in a repository |
| Where it runs | Anthropic’s service, reached from apps on web, macOS, Windows, iOS, and Android | Your machine, in a terminal, in the directory you start it in — plus a hosted web version, a desktop app, and IDE extensions |
| What it can touch | What you paste or upload, and the tools and connectors you enable | The files in your project, your shell, git, your tests, MCP servers you connect — with a permission model deciding what runs without asking |
| How you talk to it | Conversation; it replies | Instructions; it works, shows diffs, asks before risky actions, and reports back |
| How it’s paid for | Free tier for chat; Pro at $20 a month ($17 on annual billing); Max at $100 or $200; Team and Enterprise per seat | Included with Pro, Max, Team, and Enterprise, not with the free plan; or pay per token through the Console API, Amazon Bedrock, Google Cloud, or Microsoft Foundry |
| Use it when | You want an answer, a draft, an explanation, or a review of something you paste in | You want the change made: a bug fixed across files, tests written and run, a refactor carried through |
Prices from claude.com/pricing and Anthropic’s support pages; access rules from the Claude Code setup docs. Checked September 2026.
What “on a machine” actually means.
The practical difference is not intelligence — it is the same model on both sides — but the surface. Claude the app has no filesystem: you carry context to it. Claude Code has one, and it is yours. It reads CLAUDE.md from the project root for standing instructions, keeps a transcript of every session under ~/.claude/projects/, remembers which commands you allowed, and resumes a conversation with claude --continue. All of that is state on the machine where it ran.
This is also why the permission model matters more in Claude Code. In Claude, a wrong answer is text you ignore. In Claude Code, a wrong action is a file changed or a command run, so the CLI has modes — Manual, accept-edits, plan, auto — and a settings file where you list what may run without asking. Read the Claude Code overview once and the split stops being confusing: one is a place to think, the other is a worker on a box.
Common cases, sorted.
The question people are usually asking, and which product answers it.
I want to learn or plan before touching code
Claude. Paste the design, ask the questions, get the tradeoffs. Then take the plan to Claude Code — or start Claude Code in plan mode, which reads the repository and proposes without editing anything.
I want a change made in my repository
Claude Code. The whole point is that it can open the files, run the tests, and commit. Doing this by copying files into a chat window is the thing Claude Code was built to replace.
I don’t have a terminal
Claude, or Claude Code’s desktop app and web version. The web version at claude.ai/code runs in Anthropic’s cloud against a repository you connect; the desktop app bundles the CLI. Both need a paid plan.
I’m paying for Pro or Max already
Both. The subscription covers the apps and Claude Code. Whether the plan is enough for agent work is a separate question, covered at /is-claude-code-free and /claude-code-max-plan.
The machine can be anywhere.
Because Claude Code runs on a machine rather than in a chat window, the machine becomes the interesting variable. It can be the laptop in front of you, and for most people it is. It can also be a workstation at the office, a Mac mini under a desk, or the build box with the GPU — wherever the repository, the toolchain, and the credentials already live. Anthropic’s own Remote Control and web version cover part of this; the rest is the ordinary problem of reaching a machine you own without exposing it.
That is the gap spawnd fills: one daemon on each host you own, dialing out so nothing listens, and any browser — including the one on your phone — as the console. Claude Code keeps running on the host as a persistent session; you keep watching it from wherever you are. How that works in practice — the phone included — is at /claude-code-remote.
Start
Run Claude Code on the machine that has the code.
curl -fsSL https://spawnd.dev/install.sh | shQuestions
- Is Claude Code just Claude in a terminal?
- No. It is a separate agentic tool built on the same models: it reads and edits files, runs commands, and manages git on the machine where it runs. Claude the app has no access to your filesystem.
- Do I need a separate subscription for Claude Code?
- No. Pro, Max, Team, and Enterprise plans include Claude Code. The free plan does not; the alternatives are pay-per-token access through the Console API or a cloud provider such as Amazon Bedrock.
- Can Claude Code run without my laptop?
- Yes, if it runs on a different machine. Sessions are stored on the host that runs them. spawnd keeps that host reachable from any browser and keeps each session alive in its own worker process, so the laptop can close.