Claude Code Review 2026

What is Claude Code?

Anthropic's own cost guide says the average enterprise developer burns about $13 of Claude Code usage per active day, and 90% stay under $30. That figure says a lot: people leave this tool running on real work for hours. This Claude Code review and pricing breakdown is written for Indian developers, agency leads and founders who want to know which plan to buy, what the limits feel like, and where Cursor or GitHub Copilot might suit them better.

What actually happens when you type claude inside a repo

You install a small binary, open a project folder and run claude. From there it behaves like a terminal agent rather than an autocomplete box. It reads files, searches the codebase, proposes edits across several files, runs your test command, reads the failures and tries again. It can stage changes, write the commit message and open a pull request. That loop is what Anthropic means by agentic coding, and it is the main reason to pick this tool over a chat window.

The terminal is the home surface, but not the only one. The docs list a VS Code extension (it also installs in Cursor), a JetBrains plugin, a desktop app with a Code tab, a browser version at claude.ai/code that runs tasks in Anthropic-hosted VMs, and an @Claude mention inside Slack. The VS Code Marketplace listing shows more than 28 million installs, and every surface shares one engine and one set of project files.

The models behind it are Anthropic's current line: Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1. You switch with /model. Anthropic's advice is to keep Sonnet as the everyday default and save Opus for hard architectural work, because Opus drains your allowance faster.

CLAUDE.md, hooks and subagents change how a team uses it

• CLAUDE.md is a markdown file in the project root that the agent reads at the start of every session. Put your stack, folder conventions, test command and "never touch this" rules in it. Anthropic suggests keeping it under 200 lines, since every line costs tokens on every session.
• Hooks run your own shell scripts before or after an action. A common setup formats files after each edit or trims a 10,000-line test log down to the failing lines before the model sees it.
• Subagents split work. A lead agent hands a noisy job, such as running a test suite or reading docs, to a helper that returns only a summary, which keeps the main context clean.

Around these sit slash commands and skills (packaged workflows such as a team's /review-pr), plan mode (Shift+Tab, where the agent explores and proposes an approach before writing anything), and MCP servers that connect it to Jira, Google Drive, Slack or your own internal APIs. For teams that want to build their own bots on the same engine, the Claude Agent SDK exposes the tools and permission system as a library.

Claude Code review and pricing in one table

There is no separate Claude Code subscription. It comes bundled with paid Claude plans, and the pricing page confirms the Free plan does not include it. Anthropic shows USD only on its pricing page; the rupee figures below are approximate conversions at about ₹94.7 per dollar, before GST and any card forex markup.

The other route is a Claude Console account billed per token. On current API rates, Sonnet 5.5 costs $2 input and $10 output per million tokens, and Opus 5.5 costs $4 and $20. The API credits bundled with Max are meant for building on the Claude Platform; Anthropic's help centre says they cannot pay for Claude Code or Claude app usage.

Claude Code pricing, Pro vs Max: which tier a solo developer needs

Pro at $20 suits someone who codes with the agent for a focused block each day: fixing bugs, writing tests, refactoring a module. Anthropic's help article says both Pro and Max have a five-hour session limit and a weekly limit, and it does not publish exact figures. So a long Opus session on a large repo may reach the ceiling well before the window resets.

Max 5x at $100 gives five times Pro's usage, which cuts down the waiting for someone coding with it all day. Max 20x at $200 makes sense for people running several sessions in parallel or leaving background agents on long migrations. Max also carries a separate weekly cap for Fable, while Pro reaches Fable only through usage credits. One rough rule: if you hit the Pro cap three or more days a week, Max 5x is worth pricing out.

Why the five-hour window runs out sooner than you expect

The limit is shared with claude.ai chat and with IDE use, so a morning of chat plus an afternoon in the terminal both draw from one pool. The bigger cause is context. Every request resends the whole conversation, so a session left open all day turns a one-line question into a large request. Subagents and parallel agents each send their own requests, and the docs say the experimental agent teams feature uses roughly 7x the tokens of a normal session when teammates run in plan mode. The fixes are dull but effective: run /clear between unrelated tasks, keep Sonnet as default, and check /usage, which on paid plans shows which skills, subagents or MCP servers are eating your allowance. When you do run out, you can wait for the reset, turn on usage credits billed at API rates, or upgrade.

Subscription login or API key: a billing trap to avoid

Either works. Log in with a Pro, Max, Team or Enterprise account and usage comes out of the plan. Sign in with a Console account and you pay per token. The catch, spelled out in Anthropic's help centre, is that if an ANTHROPIC_API_KEY environment variable is set and you approve it, the tool uses the key instead of your subscription and you get API charges. A key left in your shell profile can trigger this. Companies can also run it through Amazon Bedrock, Google Cloud or Microsoft Foundry.

Claude Code vs Cursor for editor-first developers

Cursor is an AI code editor; Claude Code is an agent you can bolt onto any editor. If you like inline tab completion and visual diffs inside one app, Cursor feels more natural. Cursor's current plans start with a limited free tier, then Plus at $8/month (its own Composer and Grok models only) and Premium at $30/month (≈₹2,841), which adds cloud agents and third-party models, including Claude, at API list price through top-ups. Anthropic's tool gives you its models inside a flat subscription and deeper scripting: it pipes into shell commands and runs in CI pipelines. The two can also be combined, since the extension installs inside Cursor.

Claude Code vs GitHub Copilot on price and scope

GitHub Copilot is the cheaper start. Its Free plan gives 2,000 completions and 50 chat requests a month, and Pro costs $10/month (≈₹947) with 1,500 AI credits. Pro+ at $39 (≈₹3,693) adds third-party agents, Claude among them, in preview. Copilot's strength is autocomplete everywhere plus a cloud agent tied tightly to GitHub issues and PRs. Anthropic's agent offers no inline tab completion; it is built for multi-step tasks. For a team already paying for GitHub Enterprise, Copilot is the easy yes. For heavier refactors and long autonomous runs, Claude Code is the more natural fit, though it costs at least twice as much to start.

Is there a Claude Code free alternative worth using?

Not from Anthropic: the cheapest legal way in is Pro, or a Console account with prepaid credits. If you cannot pay yet, GitHub Copilot Free includes agent mode and a CLI within its monthly limits, and Cursor's free tier gives limited agent use. Fine for learning, but none offers Anthropic's top models with generous limits for free.

Permission prompts and sandboxing on a client codebase

In manual mode the tool starts read-only and asks before editing files or running commands, and commands like curl are not auto-approved. Auto mode, the default starting mode for interactive terminal and VS Code sessions, has a separate classifier model approve or block actions instead of you; your allow and deny rules still apply and organisations can switch it off. Optional sandboxing isolates bash commands at the file system and network level (not on native Windows, only WSL 2). Anthropic is clear that you remain responsible for reviewing what you approve, and that prompt injection is reduced, not eliminated.

What Anthropic keeps from your sessions

On Free, Pro and Max, your code and prompts can be used for training if the privacy setting is on; with it on, data is retained for five years, and with it off, 30 days. Team, Enterprise and API usage falls under commercial terms with no training by default and 30-day standard retention, with zero data retention available to qualifying Enterprise accounts. Transcripts are also stored locally in plaintext under ~/.claude/projects for 30 days. Telemetry can be disabled with environment variables. Note that the GitHub repository is public for issues and plugins, but the licence is proprietary, not open source.

Where the terminal learning curve bites

Developers who live in VS Code may find the CLI unfamiliar at first: approvals, slash commands and context management take a week to become habit. The bigger drawbacks are commercial. Limits are not published as numbers, so planning is guesswork. Heavy users are pushed toward $100 or $200 a month, which is real money in rupees. Anthropic's pricing page does not list UPI as a payment option, and there is no free tier to try it on.

A sensible first week with Claude Code

Install with the native script (curl on macOS, Linux and WSL; a PowerShell one-liner on Windows), or use Homebrew or WinGet. You need macOS 13+, Windows 10 1809+, or a recent Ubuntu, Debian or Alpine, and 4 GB of RAM. Start on Pro monthly. Run it first on a side project, ask it to explain the codebase, then ask it to draft a CLAUDE.md and trim that file by hand. Try plan mode before any multi-file change. Watch /usage for a week: if you keep hitting the cap, move to Max 5x; if you barely touch it, Pro is enough.