The state of play
OpenCode has become the most widely used open source AI coding agent, and a July 20 profile of the project put its trajectory in sharp relief. The terminal-based agent now reports around 8 million monthly active users and more than 172,000 GitHub stars, numbers that make it one of the fastest-growing open source developer tools on record. It reached that scale roughly a year after a public launch in June 2025, growing through developer word of mouth rather than a marketing budget. For engineering leaders weighing how AI coding tools enter their organizations, OpenCode's rise is a signal that the open source option has moved from a curiosity to a default many developers already run.
The project comes from Anomaly, the company built by the team formerly known as Serverless Stack, or SST. Founders Jay V, the chief executive, and Frank Wang, the chief technology officer, are University of Waterloo alumni who previously took SST through Y Combinator, raised about 1 million dollars after demo day, reached 25,000 GitHub stars, and turned profitable by 2025. That history matters because it shows a team that has shipped and monetized open source infrastructure before. OpenCode itself carries more than 900 contributors and over 13,000 commits, a level of community activity that few venture-backed proprietary tools can claim, and that breadth of contribution is part of why the project keeps pace with commercial rivals.
The model-agnostic bet
OpenCode's defining design choice is that it stays independent of any single model provider. The agent works with more than 75 large language model providers, including Anthropic's Claude, OpenAI's GPT family, Google's Gemini, DeepSeek, and local models run through Ollama. Jay V describes OpenCode as "a product designed to use AI," and says the company is not betting on any single model or provider winning. That positioning is a direct answer to the risk enterprises cite most about AI coding tools, which is lock-in to a vendor whose model, pricing, or availability can change without warning. A developer can point OpenCode at whichever model performs best for a task or fits a compliance constraint.
For a CTO, model independence changes the procurement calculus. A team standardizing on a proprietary agent tied to one lab inherits that lab's roadmap and pricing power, and switching later means retraining developers on a new tool. OpenCode decouples the interface from the model, so an organization can negotiate model access separately from the agent its developers use daily, and it can route sensitive code to a local model while sending other work to a hosted frontier model. That flexibility also protects against the fast churn in model quality, since the best coding model this quarter may trail the next. Keeping the tool constant while the models rotate is a hedge against that volatility.
How Anomaly makes money
An open source agent still needs a business model, and Anomaly's runs through a hosted service rather than the core tool. The company projects roughly 25 million dollars in annualized revenue from a paid, hosted-model offering layered on top of the free open source agent. That structure follows a familiar open source commercial pattern, where the software stays free and inspectable while the vendor sells the convenience of managed access, billing, and infrastructure around it. The approach lets Anomaly capture revenue from teams that want a supported path without forcing the broader community behind a paywall, which in turn keeps the contribution engine that drives the project's momentum running.
That revenue figure, reached without a large sales organization, tells enterprise buyers something about demand. Developers are adopting OpenCode on their own and, in meaningful numbers, paying for the hosted convenience once they rely on it. For engineering leaders, the signal is that a bottom-up tool has crossed into budget territory, and it is worth getting ahead of rather than discovering through an expense report. The healthier move is to evaluate OpenCode deliberately, decide whether the hosted service or a self-managed deployment fits the organization's security posture, and set guidance before shadow adoption sets the default. A tool this widely used inside developer teams rarely stays invisible to procurement for long.
What it means for tool strategy
OpenCode's scale forces a question that many engineering organizations have deferred: what is the standard for how AI coding agents get adopted internally. The proprietary options from major labs and well-funded startups compete on model quality and polish, while OpenCode competes on openness, breadth of model support, and the trust that comes from inspectable code. A team that values auditability and wants to avoid committing to one model vendor now has a mature, heavily used option to standardize on. The presence of a credible open source leader changes the negotiating position of any organization talking to proprietary vendors, because there is a viable alternative that costs nothing to trial.
The open codebase also answers security and compliance questions that proprietary agents struggle with. A regulated organization can read exactly how OpenCode handles code, what it sends to a model, and where data flows, which is difficult when the agent is a closed service. The project states that it stores none of the user's code, and an enterprise can verify that claim against the source rather than accepting it on faith. That transparency is increasingly the deciding factor for security teams evaluating whether developers may use an AI agent at all. For organizations in sensitive sectors, the ability to self-host and audit can outweigh a marginal quality edge from a closed competitor.
The risks to weigh
OpenCode's momentum does not erase the questions leaders should ask before endorsing it. A project driven by community contribution and a young company carries key-person and sustainability risk, and an organization standardizing on it should understand Anomaly's funding, governance, and long-term commitment. The hosted service concentrates dependency on a single vendor even though the agent is open, so teams relying on managed access should model what happens if pricing or terms change. Broad model support is a strength, yet it also means the quality of a coding session depends heavily on which model a team wires up, and getting that configuration right takes deliberate evaluation instead of defaults.
There is also the ordinary discipline that any AI coding agent demands. An agent that edits and ships code needs guardrails, including review of its changes, scoped permissions, and clear boundaries on what it may touch in a repository. OpenCode's openness helps teams implement those controls, since they can see and constrain what the agent does, but the controls still have to be built and enforced. Engineering leaders should treat widespread developer enthusiasm as a reason to set policy, covering which models are approved, whether the hosted service is permitted, and how the agent fits existing code review and security gates. Adoption without that scaffolding invites the same problems any powerful tool creates.
The bottom line
OpenCode's numbers make it impossible for engineering leaders to treat open source AI coding agents as a fringe choice. Eight million monthly users, 172,000 stars, a deep contributor base, and a hosted service on track for 25 million dollars in revenue describe a tool that has already won broad developer trust and found a working business model. Its model-agnostic design speaks directly to the lock-in fears that make enterprises cautious, and its inspectable code answers the security questions that keep agents out of regulated environments. The project has earned a place on the shortlist for any organization deciding how AI enters its development workflow.
The work for leaders is to convert that developer enthusiasm into a considered strategy. That means evaluating OpenCode against proprietary agents on the dimensions that matter for the organization, which include model choice, auditability, security posture, and total cost once the hosted service enters the picture. It means deciding between self-hosting and managed access, approving a set of models, and wiring the agent into existing review and compliance controls before it becomes the default by accident. Handled that way, OpenCode gives an organization a powerful, open, and portable foundation for AI-assisted development. Ignored, it will spread through developer teams on its own terms and set the policy for you.
