OpenCode just crossed 180,000 GitHub stars and 7.5 million monthly active developers. It has now surpassed Claude Code in user count. The open-source terminal coding agent built by the SST team did this in roughly 18 months, without a marketing budget, without a paid tier anyone had to buy, and without a corporate parent. The coding tool market has a new leader — and the irony is that Anthropic and Cursor built the market for it.
What OpenCode Actually Is #
OpenCode is a terminal-native, MIT-licensed AI coding agent. The SST team — the same people behind the SST serverless framework — built it as an alternative to paying $20/month for a tool that locks you into one model and one interface.
Technically, it runs as a local HTTP server. The terminal TUI (built in Go with Bubble Tea), the desktop app (now in beta on macOS, Windows, and Linux), and the VS Code extension all communicate with that server. Your conversations are stored in SQLite on your machine. Nothing leaves unless you send it to a model API — and which API you use is entirely your choice.
That choice covers 75+ LLM providers out of the box: Claude, GPT, Gemini, GLM, Kimi, MiniMax, DeepSeek, Qwen, and local models via Ollama. You connect your own API key, pay provider rates directly, and keep the difference. The GitHub repository has full setup documentation.
The LSP Feature Most Developers Miss #
The headline feature everyone focuses on is model flexibility. The feature that actually matters day-to-day is the LSP integration.
OpenCode connects to 40+ Language Server Protocol servers out of the box. That means it feeds compiler diagnostics, type errors, and linter warnings back to the model in real time. When the model writes code that breaks a TypeScript type, it knows immediately and self-corrects — before it hands you a diff to review. Most closed-source tools still skip this feedback loop. OpenCode built it in from the start.
Why Terminal Agents Won #
When OpenCode launched in early 2025, terminal-first was considered a deliberate niche. By mid-2026, it is the mainstream. Four things converged:
Claude Code normalized the terminal. Anthropic shipped a command-line tool as their flagship product. Millions of developers who had been skeptical of terminal workflows adopted one because Anthropic’s name was on it. That trained the market perfectly for OpenCode.
Multi-machine workflows became standard. You write code on a laptop, deploy on a cloud box, run heavy jobs on a local workstation. A terminal agent runs anywhere you have a shell. VS Code extensions need extra configuration on SSH boxes. Terminal agents do not.
Vendor lock-in anxiety peaked. Claude Code’s weekly usage limit just dropped 17% — effective September 14. OpenAI pulled model access from Cursor. Every proprietary move pushed more developers toward owning their toolchain.
Open-weight models became competitive. In 2024, you needed Claude or GPT-4 for a capable coding agent. In 2026, GLM-5.3-Flash, MiMo, MiniMax M3, and DeepSeek V4 are competitive on most tasks. OpenCode routes to all of them. Cursor and Claude Code lock you to one provider’s choices.
Pricing: What You Actually Pay #
Three tiers worth understanding:
Free (BYOK):$0/month. Connect your own API keys from any of 75+ providers at provider rates with no markup. With Ollama and local models, you can run a capable agent for nothing.Zen (Pay-As-You-Go): Start with a $20 balance. Zero markup on token costs — pure pass-through pricing. Includes genuinely free models: Big Pickle, MiMo V2 Pro Free, Nemotron 3.5 Lightning Free. Details in theZen documentation.Go ($10/month): 24 models included — Grok 4.6, Kimi K3, DeepSeek V4 Pro and Flash, MiniMax M3, GLM-5.3 and variants, Qwen3 series. Usage limits: $30/week, $60/month. See theGo plan pagefor the full model list.
At team scale: 10 developers on Go costs $100/month versus $200/month for Claude Code Pro and $200/month for Cursor Pro.
When to Use OpenCode (and When Not To) #
OpenCode wins in SSH and remote environments, when you want to control API spend, when privacy matters (code stays on your machine in SQLite), and when you need to test multiple models against the same task.
Claude Code still wins when you need peak SWE-bench performance — 87.6% — for large autonomous refactors where model quality is the bottleneck. Cursor still wins if you want a full IDE experience with visual diffs and inline completions inside VS Code.
A Pragmatic Engineer survey of 15,000 developers in February 2026 found that 70% run two to four AI coding tools simultaneously. The question is not which one wins — it is which one for what.
Getting Started #
Install via npm:
npm install -g opencode-ai
Or run without installing:
npx opencode-ai
Inside the TUI, run /connect
to link a provider and /models
to see what is available. LSP integration activates automatically for supported project types.
The Real Story #
OpenCode is not the end of paid coding tools. Claude Code and Cursor are not going anywhere. What OpenCode ends is the assumption that you must pick one expensive tool, accept its rate limits, and live with its model choices.
The multi-tool era is here. OpenCode is the $0 (or $10) foundation of that stack — the agent you run everywhere else, on everything else, with every model you want to try. That is a genuinely useful thing that did not exist 18 months ago.
180,000 stars and 7.5 million developers clearly agree.