OpenCode Crosses 160K GitHub Stars — The Open-Source AI Coding Agent That Beat Claude Code on Adoption
OpenCode, the open-source terminal AI coding agent built by the SST team, has surpassed 160,000 GitHub stars and 7.5 million monthly active developers — making it the most-starred open-source AI coding tool in history. It runs on 75+ LLM providers and lets users bring their own subscriptions, which may explain why it grew faster than any proprietary alternative.
OpenCode has crossed 160,000 GitHub stars and 7.5 million monthly active developers, cementing its place as the most-adopted open-source AI coding agent ever built. The project hit that milestone roughly one year after launching in June 2025 — an adoption curve that outpaced proprietary alternatives including GitHub Copilot, Claude Code, and Microsoft’s MAI-Code-1-Flash.
The tool was built by the team behind SST (Serverless Stack), a framework with a long track record of developer-first infrastructure tooling. Founders Jay, Dax Raad, and Adam Elmore shipped a Go-based CLI with a Bubble Tea terminal UI that runs inside the developer’s own environment, connecting to whichever LLM the developer already pays for. That last point drives most of the adoption: OpenCode supports 75+ providers — Claude, GPT-5.5, Gemini, Grok, Mistral, and any local model via Ollama. If you have a GitHub Copilot subscription, you can route it through OpenCode. Same for ChatGPT Plus. The tool acts as an interface, not a vendor.
Under the hood, OpenCode uses a client/server architecture. The server handles model communication, tool calls, and file operations. The client is the TUI. LSP integration is built in — OpenCode feeds compiler diagnostics, type errors, and lint warnings directly back to the model, so the agent can see what the IDE sees without being the IDE. That feedback loop makes it significantly more effective on real codebases than agents that only see file contents.
Version 1.16.2 shipped on June 5. The project has had over 816 releases and 13,969 commits since launch — a pace that reflects an unusually active open-source maintenance culture. The team has kept the codebase small and the dependency surface minimal, which means it installs and runs without friction on essentially any developer machine.
The privacy argument is also meaningful. OpenCode stores no code on its own infrastructure. It never had to, because it never had infrastructure in the first place. The model calls go directly from the developer’s machine to whichever API provider they’ve configured. For teams under strict data handling rules — financial services, healthcare, defense contractors — that’s a real selling point that no hosted agent can match at the same price point (free).
The broader implication is structural: the AI coding layer is commoditizing faster than anyone anticipated. When an open-source tool with no marketing budget outgrows products backed by Microsoft and Anthropic, the differentiation story shifts from “which AI coding tool has the best model” to “which has the best IDE integration, context management, and tool-use architecture.” OpenCode’s bet is that an open, model-agnostic client wins that race by definition.