OpenCode: A Terminal-Based AI Agent with Real-World Usage Data

Indie Hacker News2026-08-30go watch the original →

the gist

OpenCode is an open-source, terminal-based AI coding agent that decouples the agent from the model provider, allowing users to leverage existing subscriptions while providing transparent, real-time telemetry on model usage.

The Shift in AI Coding Architecture

OpenCode functions as a terminal-based AI coding agent that intentionally breaks the coupling between the agent and the model provider. Unlike proprietary solutions like Cursor, Claude Code, or GitHub Copilot, which bundle the agent and the model into a single subscription, OpenCode acts as a client that connects to over 75 providers via the models.dev catalog. This allows developers to use existing credentials, such as GitHub Copilot or ChatGPT Plus, to power their coding sessions without paying redundant monthly fees.

Real-World Model Usage Insights

The most significant feature of the project is the public dashboard at opencode.ai/data, which provides a live, telemetry-based scoreboard of model usage. Contrary to industry marketing, the data shows that developers are heavily favoring models from Chinese labs. As of the report, GLM 5.3 Flash is the most used model, followed by DeepSeek and Xiaomi's Mimo. OpenAI models do not appear in the top 10 by usage volume. DeepSeek currently accounts for 74% of total token share, with China representing 23% of total traffic compared to 13% from the United States.

Operational Mechanics and Sustainability

Built by the team behind the SST framework, the project is written in TypeScript and distributed via standard package managers including npm, Homebrew, and Nix. The project maintains a high release cadence, shipping eight versions in the first three weeks of August. Notably, the project's own AI agent is the third most prolific committer to the repository, actively filing bug fixes and database migrations. The company sustains the project through a tiered model: a $10 monthly "Go" plan for open-weight models, a "Send" pay-as-you-go option for frontier models with zero markup, and an enterprise tier for organizations requiring compliance support.