Week 12 of my “I Used It for a Week” series. Last week I reviewed Aider — the terminal purist’s AI coder. This week: the open-source project that exploded to 95K+ GitHub stars and is directly challenging Claude Code for the terminal AI coding crown.

OpenCode is what happens when the open-source community decides Claude Code is too expensive and too locked-in. It’s a terminal-based AI coding agent that supports 75+ LLM providers, has a polished TUI (terminal UI), and recently launched a desktop app. The community is massive — 2.5M+ monthly developers and growing.

How It Works

npx opencode

One command. OpenCode detects your project, reads the codebase, and drops you into a terminal UI that’s significantly more polished than Aider’s chat interface. You get a split view: your conversation on one side, file changes on the other. It feels like a proper application, not a chat prompt.

You can also use the desktop app (Electron-based) or connect it to VS Code as an extension. But the TUI is where most developers live.

What Blew Me Away

The TUI is beautiful

I don’t say this about terminal apps often. OpenCode’s interface has syntax-highlighted diffs, a file tree, session history, and keyboard shortcuts that feel native. After Aider’s raw chat interface, OpenCode feels like jumping from Vim to Neovim — same philosophy, much better experience.

Provider flexibility

OpenCode connects to everything: Claude, GPT, Gemini, DeepSeek, Mistral, Ollama, LM Studio, Azure, Bedrock, and any OpenAI-compatible endpoint. The “Zen” mode gives you access to pre-tested models that OpenCode has benchmarked specifically for coding — so you don’t have to figure out which model works best.

Session management

Every coding session is saved and resumable. You can start a task on Monday, close your terminal, and pick up exactly where you left off on Tuesday. The session includes the full conversation, file changes, and context. This is something Claude Code does well too, but OpenCode’s implementation feels more robust.

LSP integration

This is the technical differentiator. OpenCode hooks into the Language Server Protocol of your project — the same system your IDE uses for go-to-definition, find-references, and error checking. This gives the AI compiler-level understanding of your code, not just text pattern matching.

In practice, this means fewer hallucinated function names, better understanding of types, and more accurate refactoring across files.

What Frustrated Me

Slower than Claude Code

For equivalent tasks, OpenCode takes noticeably longer than Claude Code. Part of this is the provider abstraction layer — routing through multiple possible backends adds latency. Part of it is that Claude Code has deep optimizations for Anthropic’s models that a generic tool can’t match.

Configuration complexity

OpenCode is highly configurable — which means there’s a lot to configure. Model selection, provider setup, permission policies, tool access, MCP servers. The defaults are sensible, but if you want to optimize your setup, expect to spend an hour reading docs.

Desktop app is early

The Electron desktop app launched recently and it shows. It works, but it’s not as stable as the TUI. Window management is quirky, and it occasionally loses connection to the backend. The TUI is the mature product; the desktop app is a beta.

The Cost

OpenCode itself is free and open-source. You pay for the LLM provider:

Setup	Monthly cost
OpenCode + Claude Sonnet	$40-80 (API usage)
OpenCode + DeepSeek V3	$5-15
OpenCode + Ollama (local)	$0
OpenCode Zen (managed)	$20/month + API costs

The Zen tier adds managed model access and benchmarked configurations. Worth it if you don’t want to manage API keys yourself.

OpenCode vs Claude Code vs Aider

Feature	OpenCode	Claude Code	Aider
GitHub stars	95K+	N/A (closed)	39K+
Models supported	75+ providers	Claude only	Any OpenAI-compatible
UI	Polished TUI + desktop	Terminal chat	Basic terminal chat
LSP integration	✅ Deep	❌ No	❌ No
Git integration	✅	✅	✅ Best
Autonomy level	High	Highest	Medium
Cost	Free + API	$100-200/mo (Max plan)	Free + API
Best for	Flexibility + polish	Complex autonomous tasks	Simplicity + value

My Verdict After 7 Days

OpenCode is the most complete open-source AI coding agent available. The TUI is genuinely good, the LSP integration gives it a technical edge, and the provider flexibility means you’re never locked into one vendor.

But “most complete” comes with complexity. If you just want to chat with AI about your code, Aider is simpler. If you want the most capable autonomous agent and don’t mind paying, Claude Code is still ahead on raw capability.

OpenCode sits in the middle — more polished than Aider, more flexible than Claude Code, and backed by a community that’s iterating fast.

Best for: Developers who want a polished open-source tool with maximum flexibility. Teams that need provider choice for compliance or cost reasons.

Not for: Developers who want the simplest possible setup (use Aider). Developers who want maximum autonomy and don’t mind vendor lock-in (use Claude Code).

Rating: 8/10 — The best open-source AI coding agent. The community momentum is real, and it’s improving weekly.

Next week: I Used Google Antigravity for a Week — Google’s agent-first IDE that claims to replace your entire development workflow.

FAQ

Is OpenCode worth using?

Yes — OpenCode is the most polished open-source AI coding agent available, with a beautiful TUI, LSP integration for compiler-level code understanding, and support for 75+ LLM providers. It’s free to use and you only pay for the API provider you choose, making it an excellent option for developers who want flexibility without vendor lock-in.

Is OpenCode better than Aider?

OpenCode has a more polished interface, deeper LSP integration, and session management that lets you resume work across days. Aider is simpler to set up, has better git integration (every change is an auto-commit), and is more lightweight. Choose OpenCode for a richer experience with maximum provider flexibility; choose Aider for simplicity and git-native workflows.

Does OpenCode work with Ollama?

Yes — OpenCode supports Ollama along with 75+ other providers including Claude, GPT, Gemini, DeepSeek, LM Studio, and any OpenAI-compatible endpoint. Running a local model via Ollama makes OpenCode completely free to use, though cloud models like Claude Sonnet will produce higher-quality results for complex tasks.

Related: How to Choose an AI Coding Agent · AI Coding Tools Pricing