opencode

OpenCode is an open-source AI coding agent. From its terminal interface, it can explain unfamiliar code, plan a change, edit project files, run commands, and help you review or undo its work. You can also use OpenCode through its desktop app or IDE extension.

Choose the model behind the agent

OpenCode separates the coding agent from the model that answers it. Connect a provider, select a model, and change models for different tasks. Its integrations cover more than 75 providers, and it can connect to local model servers such as Ollama.

Anthropic models are available through OpenCode’s Anthropic provider and its optional OpenCode Zen catalog. The current Zen list includes Claude Opus 5.5 and Fable 5.1 for demanding work, Sonnet 5.5 for everyday coding, and Haiku 4.5 for lighter tasks. Model access depends on the provider, credentials, and current catalog. Anthropic’s Claude Pro and Max subscriptions cannot be reused through third-party plugins in OpenCode; use an Anthropic API account or another supported provider for Claude models.

OpenCode can also use OpenAI models, including Codex-tuned models in the Zen catalog, as well as Google and local models. This makes it useful when you want to change model providers without changing your coding-agent interface. Remote models process prompts and project context through their provider. With a local model server, inference can stay on your machine, though setup and hardware needs depend on the model.

Main features

• Explore and edit repositories: Ask questions about project files, plan a feature, make changes across files, and run development commands.
• Choose how much control to give the agent: Use a read-only planning mode for analysis or a build agent for development. Permission rules can ask before actions, allow them, or block them.
• Switch model providers: Select among supported hosted APIs or configure a local model endpoint. Keep in mind that provider access, model quality, quotas, and billing are separate from OpenCode itself.
• Connect tools with MCP: Add local or remote Model Context Protocol servers to make external tools available to an agent.
• Customize project workflows: Use project instructions, custom agents, plugins, commands, and other configuration to adapt OpenCode to a codebase.
• Use it from scripts: Run coding tasks from the command line or connect clients to OpenCode’s server interface.

Linux support and things to consider

OpenCode provides a Linux install script, Homebrew packages, and Arch Linux packages. Its desktop app is in beta and offers Linux .deb, .rpm, and AppImage downloads. The terminal interface is the most established Linux entry point.

OpenCode’s MIT license applies to the software, not to the models or providers you connect. API access can add usage costs, while local models require suitable hardware and may need extra configuration. Anthropic model subscriptions have an additional restriction: you should not to connect Claude Pro or Max through third-party plugins.

How it compares with ChatGPT, Codex, and Claude Code

ChatGPT is a general-purpose assistant, while OpenAI Codex is the closer coding-agent comparison. OpenCode’s OpenAI provider supports ChatGPT Plus or Pro sign-in as well as API credentials, and its Zen catalog lists Codex models. Codex itself is OpenAI’s first-party coding agent across terminal, desktop, IDE, and cloud workflows. Choose Codex when you want OpenAI’s integrated agent experience; choose OpenCode when switching between providers or using local models matters more.

Claude Code is the closest match for a terminal-first workflow built around Claude. It offers Anthropic’s coding agent and model integration, while OpenCode gives you a broader choice of model providers and local inference. That flexibility comes with more provider setup and separate account or billing choices.

OpenCode suits Linux developers who want an extensible coding agent and value model choice. If you prefer a tightly integrated experience with one model provider, Codex or Claude Code may be simpler to manage.