Claude Code
Claude Code is Anthropic's agentic coding tool that runs in the terminal, IDE, and desktop, powered by Sonnet and Opus models, with checkpointing, subagents, hooks, MCP support, and a native VS Code extension—available only through paid Claude subscriptions or API access, with no free tier.
What is Claude Code?
Claude Code is Anthropic's agentic coding tool, positioned as a terminal-based assistant that executes development tasks, explains code, and handles complex multi-file work. It is not a separate product with its own pricing: every paid Claude subscription from Pro upward includes Claude Code, and its usage draws from the same five-hour session and weekly limits as claude.ai and Claude Desktop. The Free plan does not include it, and there is no standalone Claude Code subscription.
The tool has expanded beyond its terminal origins. A native VS Code extension (beta) provides a graphical sidebar with inline diffs and plan review. JetBrains IDEs are supported, and a desktop application is available on macOS, Windows, and Linux (Ubuntu 22.04+/Debian 12+ in beta). A Chrome integration, out of preview for direct Anthropic plans, lets Claude Code drive the browser—opening tabs, clicking pages, filling forms, reading console logs, and sharing login state—to test the applications it builds. The Claude Agent SDK, formerly the Claude Code SDK, exposes the same core tools, context management, and permission frameworks for building custom agents.
A defining characteristic of Claude Code is its checkpoint system. Before each edit through the Write, Edit, or NotebookEdit tools, Claude Code saves the code state, allowing /rewind (or a double-Esc) to restore code, conversation, or both. Checkpoints apply to Claude's edits but not to subagent edits, user edits, or bash commands, and Anthropic recommends pairing them with version control rather than treating them as a replacement.
How does Claude Code work?
A user gives Claude Code a task in natural language, either in the terminal or through the VS Code extension's graphical interface. The agent explores the codebase, proposes or makes edits, runs commands, and iterates until the task is complete. Subagents can be delegated specialized work—one handling a backend API while the main agent builds the frontend—and background subagents allow Claude to continue working while they run, with permission requests surfaced in the main session. Hooks trigger automated actions at specific points, such as running a test suite after code changes or linting before commits. Background tasks keep long-running processes like dev servers active without blocking the agent.
The output is surfaced as code changes, diffs, terminal output, or pull requests. In the VS Code extension, changes appear in real time in a dedicated sidebar panel with inline diffs. In the terminal, Claude Code provides a refreshed interface with improved status visibility and searchable prompt history (Ctrl+r). Claude Code connects to external tools through MCP servers, using either the v1 runtime or the v2 runtime (built on the MCP TypeScript SDK 2.0 with the 2026-07-28 protocol revision). MCP servers from claude.ai load only when the claude.ai subscription is the active authentication method.
Use Cases
- Executing multi-file refactors and feature implementations in the terminal with autonomous agent loops
- Delegating long-running or exploratory coding tasks to background subagents while continuing other work
- Using /rewind to recover from a failed experiment without manually reconstructing prior code state, within the scope checkpoints actually cover
- Driving the browser through Claude in Chrome to test the application Claude Code builds, without switching context
- Connecting to external data sources and tools via MCP servers for context beyond the local codebase
- Building custom agentic workflows with the Claude Agent SDK for financial compliance, cybersecurity, or code debugging agents
- Running Claude Code inside JetBrains IDEs or the Linux desktop app for teams outside the VS Code ecosystem
Pricing
Pricing model: Paid
| Plan | Price | Billing | Included usage | Notes |
|---|---|---|---|---|
| Pro | $20 | per month | The lowest tier that includes Claude Code. At least 5x the Free plan's usage per five-hour rolling session, plus a weekly limit. Access to Opus, Sonnet, and Haiku models. Extra usage beyond the included allowance is billed at standard API rates. $17/month when billed annually ($200 prepaid). | — |
| Max 5x | $100 | per month | 5x Pro's usage per five-hour session, higher output limits, priority access during high-traffic periods, and a weekly limit across all models. Includes $100/month in Claude Platform API credits that do not apply to interactive Claude Code. | — |
| Max 20x | $200 | per month | 20x Pro's usage per five-hour session. Includes $200/month in Claude Platform API credits (not applicable to interactive Claude Code). The highest individual tier. | — |
| Team (Standard seat) | $25 | per seat / month | Includes Claude Code with 1.25x Pro-level usage. $20/seat/month when billed annually. Central billing, SSO, and admin controls. 2-seat minimum. | — |
| Team (Premium seat) | $125 | per seat / month | Includes Claude Code with 6.25x Pro-level usage. $100/seat/month when billed annually. | — |
| Enterprise | $20 | per seat / month | Seat price plus usage billed at standard API rates. Annual billing only. Claude Code included. Adds SSO, audit logs, SCIM, custom retention, and organization-level spend controls. | — |
Common Limitations
- Shared usage pool with claude.ai and Claude Desktop · Claude Code does not have a separate quota. Every prompt, tool call, file read, and thinking block in Claude Code draws from the same five-hour session and weekly limit as conversations on claude.ai and Claude Desktop. Heavy coding sessions directly reduce the capacity available for chat and desktop work within the same period. Anthropic publishes no token count for these limits—only multiples relative to the Free or Pro tier—and reserves the right to adjust the share each prompt consumes.
- Usage limits are opaque and workload-dependent · Anthropic publishes no fixed token or message budget for any plan. How much of a five-hour session a prompt consumes depends on message length, attachment size, conversation length, tool use (e.g., Research, web search, MCP servers), model selection, and effort level. Two developers on the same plan can burn very different amounts in the same hour. Pro and Max users who exhaust the included allowance can enable usage credits at API rates, but the five-hour and weekly resets do not change. Weekly Claude Code limits have been permanently 25% above the pre-May 2026 standard since September 14, 2026.
- Checkpoints do not cover subagent edits or bash commands · Claude Code's checkpoint system automatically saves code state before each change made through the Write, Edit, and NotebookEdit tools, allowing /rewind to restore code, conversation, or both. However, edits applied by subagents land outside the session's checkpoint scope and are not restored by rewind. Edits made by the user or via bash commands are also excluded. Anthropic recommends using checkpoints in combination with version control rather than as a substitute for it.
- Consumer training default requires opt-out · On consumer plans (Free, Pro, Max, including Claude Code used through those accounts), Anthropic offers the option to allow conversation and coding session data to be used for model training. The setting can be turned off at claude.ai/settings/data-privacy-controls, after which new conversations and coding sessions are excluded from future training—though data already included in models trained before the opt-out remains. Commercial plans (Team, Enterprise, API, third-party platforms, Claude Gov) do not train on Claude Code code or prompts by default under Anthropic's commercial terms, unless the organization explicitly opts into the Development Partner Program. /feedback, /bug, and /share reports are retained for 5 years regardless of the training opt-out setting.
- VS Code extension is in beta and version-coupled · The native VS Code extension provides a graphical interface with sidebar, inline diffs, and plan review, but it is published as a beta. Its version number matches the Claude Code version it bundles, so a feature requiring Claude Code v2.1.286 requires extension v2.1.286 or later. The extension requires VS Code 1.98.0 or higher and works with a Claude subscription or a Claude Console API key; it is not available for JetBrains IDEs, where Claude Code runs as a terminal tool or via a separate integration.
- API access is a separate billing path with different data terms · A Claude subscription enhances the Claude Code experience but does not include API access in the sense of a bundled token allowance for programmatic use. Max and Team plans include monthly Claude Platform API credits, but those credits do not apply to interactive Claude Code and do not roll over. Using Claude Code with an API key bills per token at list rates and follows API data-retention terms rather than subscription terms. The Development Partner Program training opt-in is available only on Anthropic's first-party API, not on Amazon Bedrock or Google Cloud's Agent Platform.
Key Features
- Agentic task execution in the terminal, VS Code, JetBrains, and desktop
- Checkpointing with /rewind for code and conversation state
- Parallel subagents with foreground and background execution modes
- Hooks for automated triggers at defined points in the agent loop
- Background tasks that keep long-running processes active
Claude Code vs. Cursor, GitHub Copilot, and Codex
This comparison uses official product and pricing information available as of November 2026. It focuses on the workflow shell—where each tool operates and how it gates access—rather than ranking model output quality. Cursor, GitHub Copilot, and Claude Code can expose overlapping underlying Claude models, so the differentiating factors are integration surface, plan structure, and feature boundaries, not raw model choice alone.
| Tool | Positioning | Key differences | Best fit | Pricing note |
|---|---|---|---|---|
| Claude Code Official product page Claude pricing Claude Code data usage | Anthropic's terminal-first agentic coding tool, extended to VS Code, JetBrains, desktop, and Chrome, available only through paid Claude subscriptions or API access. | No free tier and no standalone plan: entry requires Pro at $17–20/month. Usage shares a single pool with claude.ai and Claude Desktop. Checkpoints restore Claude's Write/Edit/NotebookEdit changes but not subagent edits or bash changes. The VS Code extension is in beta. Commercial plans do not train on Claude Code data by default; consumer plans require an opt-out. Chrome integration drives the browser to test built apps. | Developers who work primarily in the terminal or want agentic coding tightly integrated with a Claude subscription, and who value checkpointing, subagents, hooks, and MCP connectivity over a standalone IDE fork. | No free tier. Pro $17/mo annual or $20/mo monthly. Max 5x $100/mo. Max 20x $200/mo. Team standard $20–25/seat/mo; premium $100–125/seat/mo. Enterprise $20/seat/mo plus API-rate usage. API per MTok: Sonnet 5.5 $2/$10; Opus 5.5 $4/$20; Fable 5.1 $10/$50. |
| Cursor Cursor pricing | An AI-first code editor built on a VS Code fork, positioning an agent at the center of the development workflow. | Cursor has a Free tier and a $20/month Pro plan, so its entry path differs from Claude Code's subscription-only model. Cursor runs all models within its own editor, while Claude Code operates across terminal, IDE extensions, and desktop. Cursor's Composer and agent mode are editor-native; Claude Code's checkpoints and subagents are agent-loop-native. Cursor's privacy mode is opt-in; Claude Code's commercial plans exclude training by default. | Developers who want a standalone editor with AI built in, a free tier to start, and an editor-centric agent experience rather than a terminal-first tool. | Free tier available. Pro $20/month. Pro+ $60/month. Ultra $200/month. Teams $40/user/month. API usage billed separately. |
| GitHub Copilot Copilot plans | GitHub's AI coding assistant integrated across VS Code, JetBrains, Xcode, Neovim, the terminal, GitHub.com, and mobile, with the widest IDE and surface coverage among the three. | Copilot has a Free tier with limited completions and agent requests, and a $10/month Pro plan—the lowest paid entry among the four. Its MCP support extends to every plan including Free. Claude Code requires a $17–20/month Pro subscription at minimum and has no free tier. Copilot's strength is breadth of IDE and platform integration; Claude Code's is terminal-native agentic depth. | Developers and teams already working within the GitHub ecosystem who want the lowest paid entry point, broad IDE support, and an AI agent integrated across surfaces without switching editors. | Free tier available. Pro $10/month. Pro+ $39/month. Business $19/seat/month. Enterprise $39/seat/month. |
| Codex (OpenAI) OpenAI Codex ChatGPT pricing | OpenAI's software-engineering agent, available in the terminal, IDE, and cloud, powered by GPT-5 Codex models. | Codex is included in ChatGPT Plus at $20/month and above, and ChatGPT's Free tier includes limited Codex access. Claude Code has no free tier and requires a Claude subscription or API key. Both are terminal-capable agents, but Codex runs on GPT-5 Codex models while Claude Code runs on Anthropic's Sonnet and Opus models. Codex's cloud task execution and ChatGPT integration differ from Claude Code's checkpoint/subagent/MCP model. | Developers already using ChatGPT Plus or Codex who want a terminal and cloud coding agent without adding a separate Claude subscription. | Included in ChatGPT Plus ($20/month) and higher tiers; limited Codex access on Free. API billed separately. |
Comparison checked against official product information on 2026-11-03. Features and pricing may change.
Frequently Asked Questions
Is Claude Code free?
No. The Free plan ($0) does not include Claude Code. It requires a paid Claude subscription: Pro ($17/month billed annually or $20/month monthly), Max 5x ($100/month), Max 20x ($200/month), Team ($20–125/seat/month), or Enterprise ($20/seat plus API-rate usage). Alternatively, you can use Claude Code with an API key, billed per token at standard API rates.
Does Claude Code have its own usage limit, separate from claude.ai?
No. Claude Code shares the same five-hour session and weekly limit as claude.ai and Claude Desktop. Every prompt, tool call, file read, and thinking block in Claude Code draws from the same plan allowance. Heavy coding sessions reduce the capacity available for chat and desktop work in the same period. Anthropic publishes usage as multiples (e.g., Pro is at least 5x Free per session; Max 20x is 20x Pro), not as token counts.
What does Claude Code's checkpoint system actually restore?
Checkpoints automatically save code state before each edit made through the Write, Edit, and NotebookEdit tools. /rewind (or a double-Esc) lets you restore code, conversation, or both to a prior checkpoint. Checkpoints do not cover edits made by subagents, edits you make yourself, or changes from bash commands. Anthropic recommends using checkpoints together with version control, not as a replacement for it.
Does Anthropic train on my Claude Code sessions?
It depends on your plan. On consumer plans (Free, Pro, Max), Anthropic offers the option to allow Claude Code session data to be used for model training; this can be disabled at claude.ai/settings/data-privacy-controls. On commercial plans (Team, Enterprise, API, third-party platforms, Claude Gov), Anthropic does not train on Claude Code code or prompts by default under its commercial terms, unless your organization explicitly opts into the Development Partner Program. /feedback and /bug reports are retained for 5 years regardless of the training setting.
What is the difference between Claude Code and Claude in Chrome?
Claude Code is the agentic coding tool that operates on your codebase in the terminal, IDE, or desktop. Claude in Chrome is a browser integration that lets Claude Code drive your browser—opening tabs, clicking pages, filling forms, reading console logs, and sharing login state—so it can test the applications it builds without you switching contexts. It is out of preview for direct Anthropic plans and is not available on Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry.
Alternative Tools
-
Cursor Cursor is an AI-first code editor built on a VS Code fork with agent mode, Tab completions, and cloud agents. -
OpenAI Codex OpenAI Codex is an agentic coding product bundled into ChatGPT plans, spanning CLI, IDE, desktop, and cloud. Plus at $20/month is the real entry point; Free and Go are demos. Usage is credit-metered with wide ranges, ChatGPT and Codex pools merge in December 2026, and disclosed security boundary issues remain relevant for teams evaluating it for sensitive codebases. -
Bolt.new Bolt.new is StackBlitz's in-browser AI full-stack app builder. It runs a Node.js runtime inside your browser tab, installs packages, starts dev servers, and generates full-stack code from prompts—no local environment setup required. -
GitHub Copilot GitHub Copilot is an AI coding agent embedded across GitHub, IDEs, the terminal, and the web. It combines unlimited completions on paid plans with agent mode, asynchronous coding agents, and code review, but its value depends on disciplined review, awareness of IP and data-training policies, and realistic expectations about PR acceptance rates and context limits. -
Lovable Lovable is an AI full-stack app builder that turns natural-language descriptions into working web applications with frontend, backend, database, auth, and deployment. It is unusually effective at getting to a functional prototype and unusually poor at producing secure production code without human intervention; third-party scans find security issues in 99% of Lovable projects, and the managed backend creates migration friction that the code-portability narrative does not fully offset. -
Ollama Ollama is an open-source local LLM runtime and model manager that makes running open-weight models on your own hardware a single command. Local inference is free and unlimited; the costs are hardware, VRAM-gated context defaults, and a documented security surface that requires deliberate network isolation.