--- title: "Claude Code Review: Anthropic's Terminal Agent Evaluated" slug: claude-code-review-2026 description: "An evidence-based review of Anthropic's Claude Code: strengths, limits, pricing, data retention and how it compares with Cursor, Codex and Copilot." date: 2026-10-22 last_reviewed: 2026-11-02 status: published type: knowledge-base author: [adam-green, liyam-flexer] image: claude-code-review-2026 image_alt: "Cover graphic: Claude Code Review: Anthropic's Terminal Agent Evaluated, over a dark blue network of connected points." tags: [Claude Code, AI developer tools, autonomous coding agents, Anthropic, CLI tools] --- # Claude Code Review (2026): Anthropic's Terminal Agent Evaluated **In short:** Claude Code is Anthropic's agentic coding tool: it reads a codebase, edits files across a project, runs commands and tests, and connects to outside tools through the [Model Context Protocol (MCP)](../glossary.html#term-model-context-protocol-mcp). It runs in the terminal, IDEs, a desktop app and the web. Our assessment is that it suits delegated multi-file work, but usage limits and token costs need active management; we ran no benchmarks of our own. ## Quick verdict and evaluation summary This verdict is our assessment, not a measured result. Claude Code is built for delegating substantive, multi-file engineering work to an agent, and it fits developers already on a paid Claude plan or an Anthropic Console account. If your organization needs a fixed monthly cost with no usage ceiling, a flat seat-based tool is easier to predict: Claude Code's plan usage windows and token consumption scale with task size and context length. | Dimension | Basis | Finding (as of 2 November 2026) | | :--- | :--- | :--- | | **Pricing and plans** | Vendor pricing and docs | Included in Pro and Max; Console and cloud-provider use is billed per token. | | **Integration** | Vendor docs | Terminal, VS Code, JetBrains, desktop app, web; MCP for external tools. | | **Multi-file reasoning** | Vendor description; no independent benchmark verified | Anthropic describes multi-file edits and root-cause tracing; we could not confirm third-party scores. | | **Speed** | Our assessment | Planning and clarifying turns make it slower than inline autocomplete for small edits. | | **Privacy** | Vendor data-usage page | Training depends on account type; see the privacy section. | --- ## What is Claude Code? Claude Code runs in your local development environment rather than in a chat panel. Per [Anthropic's overview](https://code.claude.com/docs/en/overview), it can build features, fix bugs, work across multiple files and tools, and create commits and pull requests. ### The agentic workflow loop 1. **Context ingestion:** Reads project files and instructions such as `CLAUDE.md`. 2. **Task planning:** Breaks a request ("write tests for the auth module, run them, and fix any failures") into steps. 3. **Tool execution:** Runs shell commands, searches, file edits and MCP tool calls. 4. **Iterative verification:** Runs tests, reads errors, amends code and repeats. 5. **Permission gates:** In the default mode it asks before most edits, shell commands and network access. --- ## Reported strengths and practical drawbacks We could not locate the third-party benchmark write-ups that an earlier version of this post cited (a FastAPI root-cause test and a timing comparison with Cursor and Codex), so their figures have been removed. What the vendor documents, and what we assess: - **Multi-file work and test loops:** Documented capability: Claude Code plans across files and runs commands to verify its work ([overview](https://code.claude.com/docs/en/overview)). How well it does this on your codebase is something to test in a pilot. - **Usage limits:** On Pro and Max, Anthropic's [pricing page](https://claude.com/pricing) describes limits that apply "every 5 hours and every week". Long sessions send the whole conversation with each request, so usage can climb quickly ([cost guide](https://code.claude.com/docs/en/costs)). - **Cost on API billing:** Anthropic reports an average of about $13 per developer per active day across enterprise deployments, and $150 to $250 per developer per month, with 90% of users below $30 per active day ([cost guide](https://code.claude.com/docs/en/costs)). Treat these as vendor averages. - **Speed (assessment):** An agent that plans and asks clarifying questions will usually take longer than autocomplete on a small change. --- ## Architecture, surfaces, and deployment - **Surfaces:** Terminal CLI, VS Code extension, JetBrains plugin (it requires the CLI), desktop app, and web and mobile via claude.ai/code ([overview](https://code.claude.com/docs/en/overview)). - **Platforms:** The native installer covers macOS, Linux, WSL and Windows (PowerShell or CMD); Git for Windows is recommended on native Windows. - **Models:** As of this check, the `opus` alias resolves to Opus 5.5, `sonnet` to Sonnet 5.5 and `haiku` to Haiku 5.5 on the Anthropic API, and the default model on Pro, Max, Team, Enterprise and API accounts is Opus 5.5. It also runs through Amazon Bedrock, Google Cloud's Agent Platform (the Vertex AI integration), Microsoft Foundry and Claude Platform on AWS, where aliases can resolve to older versions ([model configuration](https://code.claude.com/docs/en/model-config)). - **Sandboxing:** Commands run on your machine under the permission mode you choose. Anthropic says `bypassPermissions` mode is meant for isolated containers and VMs only ([permission modes](https://code.claude.com/docs/en/permission-modes)). --- ## Data privacy, retention, and enterprise compliance Policy differs by account type ([data usage](https://code.claude.com/docs/en/data-usage), checked 2 November 2026): - **Team, Enterprise, API and third-party platforms:** Anthropic does not train generative models on code or prompts sent under commercial terms unless the customer opts in (for example through the Development Partner Program). Standard retention is 30 days. Zero data retention is available to qualified accounts on Claude Code for Claude for Enterprise, enabled per organization. - **Free, Pro, Max and Ultra:** You choose whether your data is used to improve future models. With the setting on, retention is 5 years; with it off, 30 days. - **Local transcripts:** Session transcripts are stored in plaintext under `~/.claude/projects/` for 30 days by default (`cleanupPeriodDays`). --- ## Claude Code vs. Cursor vs. OpenAI Codex Vendor-stated facts only, checked 2 November 2026. We have not benchmarked the three head to head, and these products change often. | Capability | Claude Code | Cursor | OpenAI Codex | | :--- | :--- | :--- | :--- | | **Interfaces** | Terminal, IDE extensions, desktop, web | IDE-based product with agents (see its site) | CLI, IDE extension, desktop app, web and cloud, SDK | | **Access and pricing** | Pro (from $17/mo annual, $20 monthly) and Max (from $100/mo) include it; Console and cloud providers bill per token | Free, Plus $8, Premium $30, Super $100, Ultra $200 per month; shared weekly usage pool; third-party models billed at usage | Included in ChatGPT plans; with an API key, billed at API prices (CLI, SDK, IDE only) | | **Tool extensibility** | MCP | See Cursor docs | See Codex docs | Sources: [Claude pricing](https://claude.com/pricing), [Cursor pricing](https://cursor.com/pricing), [Codex pricing](https://learn.chatgpt.com/docs/pricing), [Codex CLI](https://learn.chatgpt.com/docs/codex/cli). --- ## Frequently asked questions **Is Claude Code included in the Claude Pro subscription?** Yes. Anthropic's pricing page labels Pro, at $17 a month billed annually or $20 billed monthly as of 2 November 2026, as "Claude Code included". Usage limits apply every 5 hours and every week. To pay per token instead, authenticate with an Anthropic Console account or use a supported cloud provider. **How does Claude Code compare to GitHub Copilot?** Copilot began as inline code completion, and GitHub now also documents a cloud agent that can research a repository, plan changes and open pull requests. Claude Code is agent-first and works from your own terminal or IDE. The gap is narrower than it was, so test both on your own repositories. **Can Claude Code run dangerous terminal commands?** By default it asks before most edits, shell commands and network access. Other modes include acceptEdits, plan, auto (a classifier reviews actions) and bypassPermissions, which Anthropic says offers no protection against prompt injection and belongs in isolated containers only. **Does Anthropic train on my Claude Code data?** Not on Team, Enterprise or API usage unless the customer opts in. On Free, Pro, Max and Ultra accounts it depends on the model-improvement setting you choose, which also changes retention from 30 days to 5 years. Check the current policy before sending sensitive code. **Which models does Claude Code use?** As of 2 November 2026, Opus 5.5 is the default on Pro, Max, Team, Enterprise and API accounts, and you can switch with the `/model` command. Aliases for Sonnet and Haiku resolve to 5.5 versions on the Anthropic API and may resolve to older versions on cloud providers. --- ## Sources - [Claude Code overview](https://code.claude.com/docs/en/overview), Anthropic documentation. - [Data usage](https://code.claude.com/docs/en/data-usage), Anthropic documentation. - [Permission modes](https://code.claude.com/docs/en/permission-modes), Anthropic documentation. - [Model configuration](https://code.claude.com/docs/en/model-config), Anthropic documentation. - [Manage costs effectively](https://code.claude.com/docs/en/costs), Anthropic documentation. - [Claude pricing](https://claude.com/pricing), Anthropic. - [Cursor pricing](https://cursor.com/pricing), Anysphere. - [Codex pricing and plans](https://learn.chatgpt.com/docs/pricing) and [Codex CLI](https://learn.chatgpt.com/docs/codex/cli), OpenAI. - [About the Copilot cloud agent](https://docs.github.com/en/copilot/concepts/agents/coding-agent/about-coding-agent), GitHub Docs.