OpenCode: download, install and use with free and paid models

Introduction

OpenCode is a free open-source AI coding agent. It lives in your terminal (plus a desktop app and IDE extensions), reads and edits your files, runs shell commands, and works through tasks step by step.

The pricing model is simple: OpenCode itself costs nothing. You pay only for the model tokens you burn — and you can also burn zero, with free models. OpenCode talks to 75+ providers, so you are never locked to one vendor.

Free vs paid models

Two free paths, no credit card:

• Zen free models. OpenCode Zen is the team curated gateway of tested models. A rotating set is free for a limited time (feedback collection). No billing details needed for the free tier.
• Local Ollama. Run a model on your own machine — fully private, works offline. See Qwen: download, install, fine-tune and use.

Free models on Zen right now (limited-time offers, list changes):

• big-pickle
• space-bunny-free
• longcat-2.5-preview-free
• exo-free
• fledge-alpha-free
• mimo-v2.6-flash-free
• mimo-v2.5-free
• ling-3.1-flash-free
• ling-3.0-flash-fin-free
• nemotron-3-ultra-free
• nemotron-3.5-lightning-free
• muse-spark-1.3-contributor-free
• jev-1.13-free

Free-period privacy: most free models follow zero-retention, but some providers may use free-tier prompts to improve their models. Never send secrets or customer data to a free endpoint — check the Zen privacy notes first.

Paid paths, when you need maximum quality:

• Zen pay-as-you-go. Per-request billing, credits on your account, monthly limits and auto-reload available. Example prices per 1M tokens (October 2026 — verify in the docs, they move):

• Your own provider keys. Paste an Anthropic, OpenAI, Google, or any other provider key — tokens are then billed by that provider, not by Zen. The practical way to hold many models under one key is OpenRouter: a single key for 300+ models through an OpenAI-compatible API, connected the same way via /connect.

Download and install

Recommended: the install script (Linux / macOS / WSL2)

curl -fsSL https://opencode.ai/install | bash

With npm (any OS with Node.js)

npm install -g opencode-ai

With Homebrew (macOS and Linux)

brew install anomalyco/tap/opencode

On Windows: Chocolatey

choco install opencode

On Windows: Scoop

scoop install opencode

With Docker (no install at all)

docker run -it --rm ghcr.io/anomalyco/opencode

On Arch Linux

sudo pacman -S opencode

Check it works:

opencode --version

Connect a model

Start OpenCode anywhere and run /connect:

opencode

Then inside the TUI:

1. Type /connect, pick a provider.
2. For Zen: sign in at opencode.ai/auth (add billing details only for paid models), copy the API key, paste it into the TUI.
3. For a free Zen model: same flow, no billing details — just connect.
4. For Ollama: make sure ollama serve runs locally, then connect the Ollama provider.

List the models available to you:

/models (typed inside the TUI)

Pin a default model in opencode.json so every session starts with it. Minimal example (Zen model id format is opencode/<model-id>):

{ "model": "opencode/gpt-5.5" }

Use OpenCode daily

Go to your project and start:

cd /path/to/project

opencode

Let OpenCode learn the project (creates AGENTS.md — commit it):

/init (typed inside the TUI)

Core workflow:

• Ask about code. Press @ to fuzzy-find files, e.g. ask how authentication works in a specific module.
• Plan first. Hit Tab for Plan mode (no edits, only a proposal), describe the feature like to a junior developer, iterate, then Tab back to Build mode.
• Undo freely. /undo reverts the last change (repeatable), /redo brings it back.
• Share. /share creates a link to the session for your team. Nothing is shared by default.

Troubleshooting

• Command not found after install: restart the terminal (PATH update) or rerun the install script; with npm check that the global bin dir is on PATH.
• No models listed: you skipped /connect — run it and paste a key. For Ollama, confirm ollama serve and ollama list show your model.
• Free model gone: free Zen models rotate — run /models and pick another -free id, or fall back to local Ollama.
• Unexpected charges: set a monthly limit in Zen billing and review per-request usage; session-title generation uses cheap models and shows up in history.
• Weak answers: switch to a stronger paid model for hard tasks, keep a free/flash model for questions and drafts.

Official sources:

• opencode.ai
• opencode.ai/docs
• opencode.ai/docs/zen
• github.com/anomalyco/opencode
• openrouter.ai

Article author: Andrei Olegovich