OpenCode is an open source AI coding agent developed by Anomaly. From a terminal UI (TUI), you give instructions in natural language and the agent writes, edits, debugs, and explains code for you. Since its release in May 2025 it has expanded to a desktop app and an IDE extension, and its biggest strength is that it is not tied to a single AI vendor: you can swap in any of 75+ LLM providers, including local models. For people who are unsure which model to pick, the team also offers “Zen,” a gateway that collects only the models they have tested themselves.
Key Features
- An AI agent that lives in your terminal: Give instructions such as “explain this function” or “add tests” from the TUI, and the agent handles everything from reading the code to editing it. It supports fuzzy file search with
@, plus customizable themes and keybinds - Support for 75+ LLM providers: Through Models.dev you can choose from major cloud models all the way down to local models. You can also connect an existing GitHub Copilot or ChatGPT Plus/Pro account; the standard setup is to bring your own API key
- Plan mode and undo/redo: Rather than rewriting code immediately, Plan mode lets the agent lay out an approach first and then move to implementation. Undo and redo are managed by the agent as well, which makes trial and error easier
- Multi-session and session sharing: You can run several agents in parallel against the same project. Sessions can be exported as shareable links, which is useful for sharing a work log or asking for debugging help
- LSP and MCP integration: OpenCode loads Language Server Protocol data automatically so type information and errors can be passed to the LLM. Adding Model Context Protocol (MCP) servers connects external services and internal tools as extensions of the agent
- OpenCode Zen (curated model gateway): A gateway containing only models that the OpenCode team has tested and benchmarked for coding work. It gives access to 50+ models including GPT, Claude, Gemini, Grok, and DeepSeek through a single API key
Pricing
| Plan | Price | Highlights |
|---|---|---|
| OpenCode (core) | Free (open source) | Full use of the TUI, desktop app, and IDE extension. Model usage is billed by whichever provider you connect |
| OpenCode Zen | Pay as you go (per token) | 50+ verified models. Some free models are offered for a limited time. You can auto-reload $20 when your balance drops below $5, and set monthly spending limits per workspace or member |
| Enterprise | Contact for pricing | SSO and internal AI gateway integration, secure operation inside your organization. A trial is available before rollout |
Pricing is as of August 2026. Please check the official website for the latest details.
Pros & Cons
✅ Pros
- Open source and free as software, so you can adopt an AI coding agent while avoiding vendor lock-in
- Because models are interchangeable, you control the balance between cost and quality yourself, and local models are an option
- You can reuse an existing GitHub Copilot or ChatGPT Plus/Pro subscription instead of adding another one
- Plan mode, undo/redo, and multi-session support are built around the trial and error of real work
- LSP and MCP integration make it easier to delegate work that depends on type information or external tools
⚠️ Cons
- Model usage is billed separately. Only the software itself is free; your real running cost depends on the model you pick
- The high degree of configurability comes with a learning curve: choosing providers, managing API keys, and setting up MCP
- The terminal-centric interface can feel unfamiliar at first if you are used to a GUI IDE
- Automated edits carry the risk of unintended changes, so reviewing diffs in Git remains essential
- Enterprise pricing is not published, so organizations need to make a direct inquiry
Comparison with Similar Services
| Criteria | OpenCode | Claude Code | Gemini CLI | Aider |
|---|---|---|---|---|
| Provider | Anomaly | Anthropic | Community (OSS) | |
| License | Open source | Proprietary | Open source | Open source |
| Supported models | 75+ providers / local models | Claude models only | Mainly Gemini models | Broad support for major providers |
| Interfaces | Terminal / desktop / IDE extension | Terminal / IDE extension | Terminal | Terminal |
| Notable features | Plan mode, multi-session, Zen gateway | Usage included in subscription, Agent Skills | Free tier via a Google account | Editing scoped to Git commits |
Who Is It For
- Developers who want to choose models freely rather than depend on one AI vendor
- People who already pay for GitHub Copilot or ChatGPT and want to use that account in the terminal too
- Anyone working in an environment where code cannot leave the organization and local models or an internal AI gateway are required
- Developers who prefer a terminal-centric workflow and want to delegate work without leaving it
- People who find it hard to judge which model suits coding and would rather start from a verified shortlist (Zen)
Summary
OpenCode is a direct answer to the request “I want an AI coding agent, but I do not want to give up control over which model it uses.” The software itself is open source and free, you can pick from 75+ providers, and existing Copilot or ChatGPT subscriptions can be reused. In exchange you take on the model usage costs and the initial setup effort, so the more hands-on you are, the more you get out of it. A good starting point is to hand a small task to a free model or one of Zen’s verified models, see how it feels, and scale up from there.