Claude Code from Anthropic, Codex from OpenAI and Gemini CLI from Google can work on code in a repository. This comparison is for developers choosing a coding agent that can inspect files, propose edits and use development tools.

Short answer

Start with the tool your team can use under its account and data-access rules, then compare it on a representative task. All three support repository work. Their documentation does not establish that one always reasons better or handles every large codebase more accurately.

This comparison covers the named coding products and their documented workflows, checked November 1, 2026. It does not compare the general chat apps, rank underlying model versions or report hands-on speed and accuracy tests. GitHub Copilot is a separate product and is outside this comparison.

Repository access and workflow

Task Claude Code Codex Gemini CLI
Work from a terminal Inspect files, edit code and run commands Inspect files, edit code and run commands Use file and shell tools in a terminal session
Work near an editor IDE integrations are documented An IDE extension is available IDE integration is documented
Supply project instructions Repository instructions and configuration Project instructions and skills Project context through GEMINI.md
Check access and cost Review subscription or console access Review sign-in and usage terms Review Google sign-in, API key or Vertex AI access

A command-line interface, or CLI, lets you work through commands in a terminal. An integrated development environment, or IDE, combines an editor with tools such as a debugger. Choose the interface you actually use before comparing subscriptions.

Who should pick which

Claude Code: Compare it when your team already uses Anthropic and wants repository changes alongside code explanations. Review which surfaces and account types your organization permits, and check the actual diff rather than treating a convincing explanation as proof.

Codex: Compare it when you want OpenAI's coding agent in a terminal or IDE workflow. It can inspect a project, edit files and run commands. It is broader than autocomplete, and you do not need to describe it as access through GitHub Copilot.

Gemini CLI: Compare it when you want a terminal agent and your team can use Google's supported authentication options. Confirm whether your account requires a Google Cloud project and which quota or billing terms apply. A large context limit alone does not prove that a reply accounts for every file correctly.

Cost and review

Subscription access and API usage can have different billing and usage limits. Compare the current terms for the account route you intend to use, including staff seats and additional usage. No prices are quoted here, and taxes and optional services are excluded. Prices and limits change.

Give each candidate the same small change, relevant files and acceptance criteria. Review the resulting diff, run the project's checks and record what needed correction. That gives you evidence about your own codebase without turning a documentation comparison into an invented benchmark.

Verdict

Choose based on supported access, development workflow and the changes you can verify. Claude Code, Codex and Gemini CLI all support work beyond generating a single function. Keep a review step for edits and commands regardless of which agent you choose.

Updated November 1, 2026: Replaced the incorrect Codex-through-Copilot description and unsupported model rankings with a comparison of named coding products.

Sources: Anthropic, Claude Code overview. OpenAI, Codex CLI. Google, Gemini CLI documentation and authentication options. Checked November 1, 2026.