Claude Code Review

Claude Code is not a second subscription sitting next to Claude. It is included on every paid Claude plan, and it is absent on the free one, so the buying question is which seat covers a real week of agent work. A five-hour window plus a weekly cap is the limit you will hit.

Anthropic does not publish a fixed prompt count, because a migration and a one-line fix do not cost the same.

Toolradar data: the September 2026 AI coding ranking evaluated 153 tools. Claude Code and Cursor both have directory pages there.

The longer head-to-head is Claude Code versus Cursor on that site.

How we compared: pricing and product pages for Claude, Cursor, GitHub Copilot, Codex, and Devin, read on September 24, 2026, in USD. A hands-on setup guide lives at how to use Claude Code.

Claude Code is Anthropic's coding agent for the repo you already have open. It reads the repository, edits files, runs commands, and can open a pull request, so the seat pays for work on the codebase rather than another chat tab.

Anthropic's pricing comparison marks it unavailable on Free and available on Pro, both Max tiers, Team, and Enterprise, including self-serve Enterprise, so a free login cannot turn the agent on.

You can run it in the terminal, in VS Code (the extension also installs inside Cursor), and in JetBrains. It also runs in the Claude desktop app, the browser, and the iOS and Android apps. You do not have to switch editors to use it, which is the point of that spread.

A CLAUDE.md file in the project root is read at the start of every session. If the repo already carries an AGENTS.md for another agent, Claude Code can read that file on its own or beside CLAUDE.md, so existing agent rules still apply.

The product asks before it writes a file or runs a command, which is the control you want on a shared repo. Anthropic says the terminal session talks to the model without a remote code index. Set ANTHROPIC_API_KEY and the CLI skips the subscription login and asks you to approve that key.

The session then bills the Console at API rates instead of the seat. Teleport, which pulls a web session into the terminal, requires a claude.ai subscription rather than that key, so a Console-only account cannot make that handoff.

Individual plans let you opt out of model training, while Team and Enterprise do not train on your content by default, which is what to clear with counsel before a pilot. SSO and SCIM stay off Pro and Max, so a personal seat cannot meet an identity rule.

They start on Team for SSO, and on Enterprise for SCIM, audit logs, and a HIPAA-ready offering, so none of that sits on a personal seat.

Install the CLI, open a project, and run claude, then describe the change in plain language. The agent searches the tree, edits several files, and can run the test command you already use, so the check is the one the repo already trusts.

Hooks can format after an edit or lint before a commit, and skills wrap a repeatable prompt, such as a pull-request review, so the team does not retype it.

Limits reset on a rolling five-hour window, and paid plans add a weekly cap on top. Chat on the web, the desktop app, and Claude Code share that allowance, so a long agent session can pause chat until the reset.

When the bar is empty you wait for the reset, move to a higher plan, or turn on usage credits, which continue at standard API rates. There is no published message quota to budget against, so you plan the week rather than a message count.

Cloud routines keep running after the laptop sleeps, so a task can finish with the lid closed. You create them on the web, in the desktop app, or with /schedule in the CLI.

For review comments on the pull request after the agent opens it, the AI code review list is the next decision, and CodeRabbit is the dedicated reviewer rather than a second coding agent.

Claude Code has no price of its own, so you pay for a Claude plan and the agent is either included or absent. Pro is the individual entry, and Max is monthly only, in a 5x tier and a 20x tier, each measured against Pro inside the five-hour window.

A heavier week costs more per month, because you are buying a multiple of that same window. Annual billing exists for Pro and for Team, not for Max, and the annual Pro charge is $200 up front.

Team needs at least two members and stops at 150 seats, so one person cannot buy SSO here. Standard seats include 1.25 times Pro's per-session allowance, and Premium seats include 6.25 times Pro, which is five times a Standard seat. Premium is for the people who actually run long sessions.

Limits are per person, so one developer burning the week does not empty a teammate's bar. Usage credits, if an admin turns them on, let that person continue at API rates. Monthly Standard and ChatGPT Business, billed monthly, are each $25 per user, the comparison if the alternative is a ChatGPT team seat.

Premium monthly is $125 per user, and the annual Premium seat matches the Max 5x sticker, per member, so you pay that multiple once per person.

Enterprise bills differently from a flat individual seat, and the invoice grows with usage. The pricing FAQ lists a seat fee plus usage at API rates, billed annually, with SCIM, audit logs, custom retention, and role-based access.

Anthropic's own cost guide, on the Claude Code costs page, puts a typical enterprise deployment around $13 per developer per active day and $150 to $250 per developer per month. It also says costs stay under $30 per active day for 90% of users. That range is their figure, from pilots, not a quote you can paste into a budget.

On the API, Opus 5.5 lists $4 per million input tokens, and output is five times that input rate, so a long reply costs more than the prompt. US-only inference is 1.1 times the input and output rates, and fast mode doubles those Opus 5.5 token rates.

Cursor's docs recommend Pro Plus, at $60/mo, for daily agent use, the plan to compare when the agent runs all day. Teams adds $0.25 per million tokens on third-party models, including Auto routes that land on one, so the seat is not the whole team bill.

Copilot bills extra AI credits at $0.01 each, and it does not meter inline completions on paid plans. A completions-heavy day stays off the meter that agent tasks spend.

Claude Code fits a developer who already lives in the terminal or who wants the agent beside VS Code or JetBrains without adopting a second editor. Indie hackers and small teams get the cleanest bill on Pro, because one login covers chat and the agent.

Engineering leads who need SSO should start at Team, and they should assign Premium only to the people who will actually run long sessions, since the limit does not pool.

It is the wrong default when the daily habit is Tab-to-accept completions. Copilot keeps those completions off the credit meter on every paid plan. Cursor's Tab model is the reason people stay in that editor, and the Cursor review is the editor comparison.

The Cursor versus Copilot note covers that pair without the terminal agent in the middle.

Skip a personal Max plan when the company will pay for a seat. Enterprise's seat fee looks small until API usage lands on the same invoice, which is why Anthropic publishes a per-active-day average instead of a flat seat.

If the bug is a failing test rather than a missing feature, the debugging tools list is a narrower buy than another coding subscription.

Devshot's daily brief is where a pause like Codex's September signup stop belongs. Subscribe free if you want that note in the inbox.