claude-code

Claude Code is Anthropic’s AI coding agent for working with software projects from a terminal. It can read a repository, explain unfamiliar code, edit files, run commands, and help with Git tasks such as reviewing changes or preparing commits. A prompt can describe a bug or feature in plain language, while Claude Code uses project files and command-line tools to work through it.

Claude models in Claude Code

Claude Code uses Anthropic’s hosted models. The current choices include Opus for complex reasoning and longer tasks, Sonnet for everyday coding, and Haiku for faster, simpler work. Claude Code also supports Fable for demanding, long-running agent sessions where the account and provider allow it. Model names and availability change over time, and can differ between Anthropic accounts and third-party providers.

This is not a local LLM runner: model inference requires an internet connection and sends prompts and relevant project context to the configured service. Claude Code can authenticate through a Claude subscription or Anthropic Console, and can also connect through Amazon Bedrock, Google Cloud’s Agent Platform, and Microsoft Foundry. API or cloud-provider usage may be billed separately from a subscription.

Main features

• Work across a codebase: Read and change files, search project content, run tests and other shell commands, and use Git as part of a coding task.
• Keep project guidance between sessions: Add a CLAUDE.md file with project conventions, architecture notes, and preferred commands.
• Automate repeatable workflows: Use skills and plugins for reusable instructions, and hooks to run actions around Claude Code events.
• Connect external tools: Configure MCP servers to make tools and information from other services available during coding sessions.
• Delegate coding tasks: Use subagents for focused work and the Agent SDK to build custom workflows around Claude Code.
• Use it beyond an interactive terminal session: Run prompts in scripts or CI workflows, and integrate Claude Code with supported editors.

Linux support and limitations

The terminal client supports Linux on x64 and ARM64 systems, with Ubuntu 20.04+, Debian 10+, and Alpine Linux 3.19+. Anthropic provides a native installer and signed apt, dnf, and apk repositories. Native installs update automatically; package-manager installs are updated through the system’s usual package workflow. Claude Code needs an internet connection and at least 4 GB of RAM.

A desktop app with a Code interface is also available in beta for Ubuntu and Debian. The CLI remains the most direct option for Linux users who are comfortable working in a terminal. Claude Code itself is proprietary, and it does not run Claude model weights locally.

How it compares with ChatGPT, Codex, and OpenCode

ChatGPT is a general-purpose assistant for conversation, research, and other tasks. For coding-agent comparisons, OpenAI Codex is the closer match: it offers coding workflows through its CLI, desktop app, IDE extension, and cloud environments. Claude Code is a natural fit for developers who want an agent centered on a local terminal workflow and Anthropic’s Claude models. Codex may suit people already working in OpenAI’s ecosystem or who want to delegate tasks to Codex Cloud.

OpenCode is another terminal-based option. It is open source and supports many model providers, including local models, which gives users more choice over the model backend. That flexibility also means selecting and configuring a provider is part of the setup.

Claude Code is worth considering if you want an agent to work directly with a Linux project and can send its context to a hosted model service. If you need offline inference, prefer an open-source tool, or want to switch freely among local and hosted models, OpenCode may be a better fit.