OpenCode
FreemiumThe open source AI coding agent.
Published 27 September 2026
Scores
Popularity5/5
Roughly 198K GitHub stars and 900+ contributors as of mid-2026, reported 7.5M+ monthly users — the most-starred project in the open-source AI coding agent space, ahead of every other independent (non-vendor) entrant.
Learning Curve3/5
Getting a first session running is a quick `npm install` plus an API key, but making full use of Build/Plan modes, MCP/LSP integration, and choosing among 75+ providers involves more configuration surface than a single-model vendor tool.
Flexibility5/5
Bring-your-own-model architecture, custom tools and plugins, configurable permissions/keybinds/formatters, and MCP/LSP support give it wide-open extensibility with essentially no imposed workflow.
Performance4/5
Output quality tracks whichever underlying model is configured rather than a fixed baseline — with a strong model (e.g. Claude or GPT) selected, it performs competitively with leading vendor agents, but there's no single benchmark number the way there is for a single-model product.
Portability5/5
MIT-licensed, installable via npm/Homebrew/Docker, no code stored on OpenCode's servers, and not tied to any single model provider — among the least locked-in options in the subcategory.
About OpenCode
OpenCode is an open-source AI coding agent built by the team behind the SST serverless framework (now operating as Anomaly). Released in mid-2025, it grew rapidly into one of the most-starred projects in the AI coding agent space, driven by its bring-your-own-model positioning: rather than shipping with a single default model like most vendor coding agents, OpenCode connects to Claude, GPT, Gemini, local models via Ollama, or any of 75+ providers through Models.dev, and can reuse an existing GitHub Copilot or ChatGPT Plus/Pro subscription instead of requiring a separate API key.
The agent runs primarily as a terminal UI (TUI) with a persistent SQLite-backed session store, alongside a beta desktop app and IDE extensions, so the same sessions and configuration follow a developer across surfaces. It ships two built-in agent modes: "build" (full read/write access to the codebase) and "plan" (read-only, for reviewing changes before they're applied), plus a general-purpose subagent for multi-step search and research tasks. Sessions can be shared via a generated link, and the tool integrates with LSP and MCP servers, custom tools/plugins, and configurable permissions, keybinds, and formatters.
For developers who don't want to manage their own provider API keys, OpenCode Zen is an optional managed model gateway operated by the OpenCode team: a curated, pre-verified set of 40+ models billed through a single pay-as-you-go account, plus a lower-cost OpenCode Go subscription tier for open-weight models. Neither is required to use the core agent. OpenCode itself does not store code on its own servers, and installation is available via npm, Homebrew, Docker, and standard package managers, keeping the tool itself fully self-hostable and provider-independent.
Key Features
- Model-agnostic: connects to 75+ LLM providers via Models.dev, including local models
- Reuses existing GitHub Copilot or ChatGPT Plus/Pro subscriptions instead of requiring a new API key
- Build and Plan agent modes: full-access editing vs. read-only review
- Terminal UI, beta desktop app, and IDE extensions share the same sessions
- Shareable session links via `/share`
- LSP and MCP server support, custom tools/plugins, configurable permissions
- Optional OpenCode Zen managed model gateway and OpenCode Go subscription tier
- No code stored on OpenCode's own servers, privacy-focused by design
Pros
- No single-provider lock-in — swap models per task or provider outage without switching tools
- Large, fast-growing open-source community (900+ contributors) means rapid bug fixes and feature velocity
- MIT license and self-hostable installation methods keep the tool itself fully under the developer's control
- Reusing an existing Copilot or ChatGPT subscription lowers the cost of trying it
- Consistent experience across terminal, desktop, and IDE surfaces
Cons
- Bring-your-own-model design means actual usage costs depend entirely on the provider chosen, with no single predictable price
- Younger, more fragmented ecosystem than single-vendor agents, and model quality varies by which provider is configured
- The desktop app is still in beta
- Community-run project without a single accountable enterprise vendor, which some teams weigh against vendor-backed alternatives
- OpenCode Zen and OpenCode Go add their own separate billing relationship if used, on top of understanding the core BYOK model
OpenCode Pricing
Freemium- · Free, open-source agent — bring your own API key for any supported provider
- · Reuse an existing GitHub Copilot or ChatGPT Plus/Pro subscription instead of a separate key
- · Local model support via Ollama at zero marginal cost
- · Subscription access to open-weight models (GLM, Kimi, MiniMax and similar)
- · Defined usage limits
- · Pay-as-you-go managed gateway to 40+ pre-verified models
- · Single billing relationship instead of per-provider API keys
- · Pricing at or near provider list price plus a processing fee
Tech Stacks with OpenCode
Cursor Dev Stack
DeveloperCursor as both the editor and the coding agent, with Grok as the default model and Claude, OpenAI, or Gemini available from the same model picker, plus your pick of GitHub or GitLab for version control.
Devin Dev Stack
DeveloperDevin Desktop as the editor and agent manager, with Devin Cloud for work that keeps going without your machine, Claude as the default model and OpenAI, Gemini, or Grok in the same picker, plus your pick of GitHub or GitLab for version control.
Development
LLM
Orca Dev Stack
DeveloperOrca runs several coding agents side by side, each in its own git worktree, with your pick of agents and models underneath and GitHub or GitLab for version control.
Tools Related to OpenCode
Works well with OpenCode(7)
Orca is one of the agentic dev environments that wraps OpenCode unmodified, adding parallel worktree orchestration and a native app on top of the CLI.
Paseo gives OpenCode sessions cross-device continuity through a self-hosted daemon, letting a session be steered from a phone or browser as easily as the terminal it started in.
T3 Code gives OpenCode a cross-platform GUI, git-worktree isolation, and a turn-by-turn diff viewer on top of the bare CLI.
Herdr gives OpenCode sessions a persistent, reattachable terminal environment with real-time status tracking, useful when running OpenCode unattended or on a remote box.
OpenCode documents OpenRouter among its built-in providers, a common way to drive OpenCode with a wide model choice.
DeepSeek's own docs list OpenCode among the coding agents that can point at the DeepSeek API with no code changes.
Alternatives to OpenCode(12)
OpenCode is explicitly positioned as the leading open-source, model-agnostic alternative to Claude Code — both are terminal-first AI coding agents with session management and IDE/desktop surfaces, but OpenCode lets you bring any of 75+ model providers instead of being tied to Claude models.
OpenCode is an open-source, bring-your-own-model terminal agent; GitHub Copilot is a proprietary, deeply GitHub-integrated assistant with the largest installed base in the category — alternatives representing opposite ends of the openness/integration trade-off.
OpenCode is an open-source, model-agnostic terminal agent; Codex is OpenAI's cloud-based async coding agent — alternatives in the agentic coding space with very different execution models (local BYOK sessions vs. cloud sandbox delegation).
OpenCode and Pi are both open-source, provider-agnostic terminal coding agents — OpenCode ships a fuller feature set and broader native multi-editor/desktop reach out of the box, while Pi is deliberately minimal (four core tools, ~200-token system prompt) and extension-driven.
OpenCode and Qwen Code are both free, open-source, terminal-first coding agents that accept any OpenAI-compatible provider — OpenCode has no default vendor and broader native multi-editor/desktop reach, while Qwen Code defaults to Alibaba's own Qwen3-Coder model.
OpenCode is the foundation Kilo Code is built on, but Kilo ships as its own branded product with extra layers — specialized agent modes, an Orchestrator, cloud agents, and zero-markup Gateway pricing — making the two a real choice rather than the same product twice.
Learning Resources
No resources yet — check back soon.