Claude Code in 2026: how it works, permission modes, pricing, limits
A guide to Claude Code from Anthropic's docs: how the agent works, its six permission modes, what Pro, Max, Ultra and Team cost, usage limits and privacy rules.
By Shaheer Q. · · 6 min read
Claude Code is Anthropic's AI coding agent. You describe a task in plain language, and it reads your repository, edits files, runs commands and uses git. It comes with every paid Claude plan, from $20 a month, and is not on the Free plan. Its usage comes out of the same allowance as Claude chat, which matters more for your bill than the headline price.
This is a guide built from Anthropic's documentation and pricing pages, checked on 15 November 2026. It is not a hands-on review: there are no timings, screenshots or a rating here, only what Anthropic documents. A disclosure, too: this site is built with help from Claude Code, and most commits in its repository are co-authored by it. That is a reason to be careful with praise, so this guide sticks to documented facts. Anthropic has no affiliate program, so no link here earns money.
Claude Code at a glance
| Maker | Anthropic |
| What it is | An agent that reads your code, edits files, runs commands and uses git |
| Where it runs | Terminal CLI; extensions for VS Code (and Cursor and Devin Desktop) and JetBrains IDEs; the Claude desktop app; claude.ai/code in the browser; iOS and Android; GitHub and Slack |
| Platforms | macOS 13+, Windows 10 1809+ or Windows Server 2019+ (native or WSL), Ubuntu 20.04+, Debian 10+, Alpine 3.19+; 4 GB+ RAM, x64 or ARM64; internet required |
| Default model | Opus 5.5 on Pro, Max, Ultra, Team, Enterprise and the API (November 2026) |
| Price | Included in Pro ($20/month), Max ($100), Ultra ($200), Team and Enterprise seats; or pay per token |
| Free plan | No Claude Code |
| Latest version | 2.1.325 (13 November 2026) |
How Claude Code works
You start it inside a project with claude and talk to it. It decides which files to read, proposes edits, and runs shell commands such as your test suite. Three parts of the design matter most for day-to-day use.
Project memory. A CLAUDE.md file in the repository root is read at the start of every session. It is where coding standards, commands and hard rules go: which package manager to use, what never to touch, how to deploy. The more specific it is, the less you repeat yourself in prompts.
Permission modes. Claude Code asks before it acts, to a degree you choose. Per Anthropic's docs:
| Mode | What runs without asking |
|---|---|
default (shown as Manual) | Reads only |
acceptEdits | Reads, file edits and common filesystem commands (mkdir, mv, cp and similar) |
plan | Reads; edits stay blocked until you approve a plan |
auto | Everything, with background safety checks |
dontAsk | File reads in your working directories and pre-approved tools; anything that would prompt is denied |
bypassPermissions | Everything; for isolated containers and VMs only |
Since version 2.1.283, sessions started in a terminal or the VS Code extension begin in auto mode. A permissions.defaultMode in your settings file overrides that, and non-interactive runs (claude -p) and the Agent SDK still start in default. If you want to approve every command, set permissions.defaultMode to default or start with claude --permission-mode default.
Extensions. Hooks run your own shell commands before or after its actions, for example a formatter after every edit. Skills package repeatable workflows as slash commands. MCP servers connect it to outside tools, and subagents split a task across parallel workers.
Models
The default model setting resolves to Opus 5.5 on Pro, Max, Ultra, Team, Enterprise and the Anthropic API, and on Amazon Bedrock and Google Cloud too. The /model command switches to Sonnet, Haiku or Fable for a session. Fable is never the default on any plan, and depending on your plan and seat tier its usage can bill to usage credits instead of your plan's included limits. On the pricing page, Pro pays for Fable with usage credits, while Max and Ultra can spend up to half of their weekly limits on it.
What Claude Code costs
Claude Code has no price of its own. You get it through a Claude subscription, or pay per token through the Anthropic Console. Prices from Anthropic's pricing page on 15 November 2026, before tax:
| Plan | Price | Claude Code | Usage |
|---|---|---|---|
| Free | $0 | No | |
| Pro | $20/month, or $200/year ($17/month) | Yes | More than Free; the 1x baseline for the plans below |
| Max | $100/month | Yes | 5x Pro every five hours and every week |
| Ultra | $200/month | Yes | 10x Pro every week; no five-hour limit for verified accounts |
| Team, Standard seat | $25/month, or $20/month billed annually (2 to 150 users) | Yes | More than Pro |
| Team, Premium seat | $125/month, or $100/month billed annually | Yes | 5x Standard |
How the limits work matters more than the headline price:
- One pool. Anthropic's help centre says usage on claude.ai, Claude Code and Claude Desktop "counts towards the same usage limit". A long chat in the afternoon leaves less for coding in the evening.
- Two clocks. Paid plans have a session limit that resets every five hours and a weekly limit on top. Anthropic publishes no fixed message count, because usage depends on context size, model and features.
- Overflow. When you hit a limit you can wait, upgrade, or turn on usage credits on a paid plan, and your work continues on credits instead of stopping.
- Pay per token instead. With a Console account, Claude Code bills at API prices. Opus 5.5 lists at $4 per million input tokens and $20 per million output tokens; Sonnet 5.5 at $2 and $10. Anthropic's docs put the average cost at around $13 per developer per active day, and $150 to $250 per developer per month, with 90% of users below $30 per active day.
Inside Claude Code, the /usage command shows your plan usage bars, activity stats and a breakdown of what used them, such as skills, subagents, plugins and individual MCP servers. Its dollar session cost is computed locally from token counts at list price. Anthropic calls it an estimate and says it is not relevant for billing on Pro, Max and Ultra.
| Model | paid |
|---|---|
| Paid plans from | $20/month (Pro), or $17/month billed annually |
| Notes | Not on the Free plan. Included in Pro, Max ($100/month, 5x Pro), Ultra ($200/month, 10x Pro weekly, no five-hour limit) and Team seats ($25/month, $20 annually), sharing one usage pool with Claude chat. Or pay per token via an Anthropic Console API key. |
| Checked | November 9, 2026 |
Which plan to start on
Going by the documented limits alone: start on Pro, watch /usage for a week, and move up only if you keep hitting the five-hour limit. Max is five times Pro on both clocks for five times the price, so it pays off when the waits cost you working time. Ultra removes the five-hour limit, which only matters if you run long sessions back to back. If your company pays, a Team seat keeps code under commercial terms (see below), which can matter more than the usage numbers.
Privacy and safety
On Free, Pro, Max and Ultra accounts, Anthropic trains new models on your sessions only when the model-improvement setting is on. Check it in your Claude privacy settings before pointing the agent at private code. Team, Enterprise and API accounts are under commercial terms, where Anthropic says it does not train on code or prompts sent to Claude Code unless the customer opts in. Transcripts you share with /feedback, /bug or /share are kept for five years.
On safety, the permission mode is the main control. In auto mode Claude Code runs commands with background safety checks rather than asking each time, and that is now where a terminal session starts. Pair it with a sandbox or a disposable branch for anything that can reach cloud credentials or production. If your shell has a cloud CLI logged in with broad rights, an agent that can run commands has those rights too. A named, least-privilege profile and a CLAUDE.md rule about which one to use are cheap insurance.
Claude Code vs Cursor and GitHub Copilot
The main difference is the form factor. Cursor is an AI-first code editor built on a fork of VS Code, and GitHub Copilot is an assistant for completions, chat and agent tasks inside your editor. Claude Code starts from the terminal and the task: you hand it a job and review the result. It also runs inside VS Code and Cursor as an extension, so it is not strictly one or the other.
AI code editorCursorAI-first code editor with a VS Code-compatible setup, an in-editor agent and cloud agents. AI coding assistantGitHub CopilotGitHub's AI coding assistant for completions, chat and agent tasks inside your editor.For how the two most common choices differ on agents and pricing, see Cursor vs Claude Code.
Who Claude Code suits
From what Anthropic documents, Claude Code fits developers who are comfortable handing a whole task to an agent and reviewing the diff, and who already pay for Claude or are willing to. If you mostly want inline completions while you type, an editor assistant is the closer fit. If you already use Claude chat heavily, remember the shared pool before assuming Pro is enough for both.
AI coding agentClaude CodeAnthropic's agentic coding tool for the terminal, IDE, desktop and web; included in paid Claude plans.Frequently asked questions
How much does Claude Code cost?
Claude Code is included in every paid Claude plan: Pro at $20 a month ($17 a month billed annually), Max at $100 a month, Ultra at $200 a month, plus Team and Enterprise seats. It is not on the Free plan. You can also pay per token with an Anthropic Console API key. Prices checked on 15 November 2026.
Do Claude Code and the Claude chat app share usage limits?
Yes. Anthropic's help centre says usage across claude.ai, Claude Code and Claude Desktop counts towards the same limit. Paid plans have a session limit that resets every five hours and a weekly limit on top. Paid plans can turn on usage credits to keep working past the limit.
Does Claude Code work on Windows?
Yes. It installs natively on Windows 10 1809 or later and Windows Server 2019 or later, or inside WSL. On native Windows, Git for Windows is recommended so Claude Code can use Bash; without it, Claude Code uses PowerShell for shell commands.
Which model does Claude Code use by default?
As of November 2026, the default on Pro, Max, Ultra, Team, Enterprise and the Anthropic API is Opus 5.5. You can switch with the /model command to Sonnet, Haiku or Fable. Depending on your plan, Fable usage can bill to usage credits instead of your plan limits.
Does Anthropic train on Claude Code sessions?
On Free, Pro, Max and Ultra accounts, Anthropic trains on your data only when the model-improvement setting is on. Under commercial terms (Team, Enterprise and API), it says it does not train on code or prompts sent to Claude Code unless the customer opts in.