OpenCode Hits 172K GitHub Stars and 8M Users — the Most-Starred Coding Agent, Built Free and Open in Toronto
OpenCode, an MIT-licensed terminal coding agent built by Toronto startup Anomaly, crossed 172,000 GitHub stars and 8 million monthly active users in June 2026 — one year after its debut at a 30-person DevTools Toronto meetup. It is now the most-starred open-source coding agent on GitHub, ahead of Claude Code at 122,000 stars, Gemini CLI at 105,000, and Codex CLI at 83,000.
The company expects roughly $25 million in annual revenue. The tool itself is free.
How It Works
OpenCode is a terminal agent that brings your own key. It connects to any of 75+ LLM providers through the AI SDK and the Models.dev catalog — Anthropic, OpenAI, Google, Cohere, Mistral, DeepSeek, and any OpenAI-compatible endpoint. Local models run through Ollama, LM Studio, or llama.cpp. The team maintains OpenCode Zen, a curated list of tested models for agentic coding tasks, but does not mandate any provider.
Users pay only for the tokens they consume on whichever model they choose. There is no subscription, no markup, no Anthropic-only or OpenAI-only constraint.
The GitHub star count reflects that architecture. 172,000 stars on a terminal coding agent with MIT licensing and no vendor lock-in means developers who have no financial reason to star the repo are starring it anyway.
Star Count Context
| Tool | GitHub Stars | License | Provider Lock-in |
|---|---|---|---|
| OpenCode | 172,000 | MIT | None (75+ providers) |
| Claude Code | 122,000 | Proprietary | Anthropic only |
| Gemini CLI | 105,000 | Apache-2.0 | Google (mainly) |
| Codex CLI | 83,000 | Apache-2.0 | OpenAI + Bedrock |
On Terminal-Bench 2.1, Claude Code with Opus 4.8 scores 78.9% and Codex CLI with GPT-5.5 leads at 83.4%. OpenCode’s score is model-dependent — pair it with Opus 4.8 or GPT-5.5 and you get the same underlying model capability with no subscription overhead.
The Business Model
OpenCode built its distribution by refusing to extract rent from the agent layer. The result: $25 million ARR expected from tooling, training, and enterprise features built on top of a free core. The company raised nothing publicly; the founders have not named external investors.
“We were built on the premise that most of the world had still not discovered the value of agentic coding,” founder Jay V said in a BetaKit interview this week. “We’re riding that wave now.”
The company added more than 600 contributors and 13,000 commits in the first year. It hit 650,000 monthly active users in five months, then grew to 8 million — a 12x gain on top of an already fast start.
The Structural Point
Every proprietary coding agent — Claude Code, Codex CLI, Gemini CLI — is a vendor’s distribution mechanism. They exist to route developers toward paying for that vendor’s models. OpenCode’s architecture inverts that: it optimises for the developer, not the vendor.
The timing matters. DeepSeek V4-Pro put frontier-quality output at $0.87 per million tokens in June. OpenCode paired with a cheap, capable model is now a genuinely viable production stack for cost-sensitive engineering teams who cannot absorb $500-$2,000 per engineer per month in Claude Code billing.
LogRocket’s AI dev tool power rankings placed OpenCode at number one in June 2026, displacing Cursor. The displacement signal predates the OpenAI price-cut discussions by weeks. Whatever OpenAI decides on token pricing, the open-source ecosystem built the alternative before the incumbents noticed the problem.
Numbers
- GitHub stars: 172,000 (MIT, repository anomalyco/opencode)
- Monthly active users: 8 million
- Annual revenue (expected): $25 million
- LLM providers supported: 75+
- Contributors: 900+
- Latest release: v1.16.2 (June 5, 2026)
- Languages supported: all major via LSP integration