# OpenCode Features and Practical Usage
🧘 "Which provider and model is actually good for coding?" OpenCode Zen takes that worry off your plate — an AI gateway of vetted models only.
🏷️ Title: Curated model list
🔗 URL:
📘 Overview
OpenCode Zen is an AI gateway that offers only models vetted and optimized for coding agents. Instead of "hoping a model works," the OpenCode team benchmarks model/provider combinations and talks with each provider to run them at their best. It lets people who are unsure about provider selection reach high-quality models without configuration hassle.
⚙️ How It Works
・Getting started: Sign up at add billing details, copy your API key. In the TUI run `/connect`, pick OpenCode Zen, and paste the key.
・Specifying models: Model IDs use the `opencode/
` format. For example, set `opencode/gpt-5.5` in your config file. Run `/models` in the TUI to see the full list.
・Lineup: 50+ models including OpenAI (GPT 6 and GPT 5.x), Anthropic (Claude Opus/Sonnet/Haiku), Google (Gemini), plus Qwen/DeepSeek/Kimi/GLM. Limited-time free models exist too.
・Billing: Pay-as-you-go per 1M tokens. If your balance drops below $5, Zen auto-adds $20 (configurable or disableable). You can also set workspace-wide or per-member monthly caps.
🛠️ Practical Usage
Set a Zen model as your default by adding `"model": "opencode/gpt-5.5"` to your config file.
A simple flow is to try a free model (like Big Pickle) to check behavior, then switch to a higher-quality model per task. For org use, Admin/Member roles let you centrally manage enabled models and spending limits.
💡 Use Cases
・A team with little model-selection experience starts coding with stable quality, staying within the vetted list.
・Switch between cheap and high-performance models within Zen per task to balance cost and quality.
・Bring your own OpenAI/Anthropic key (BYOK): billing goes directly to the provider while model management stays unified in Zen.
⚠️ Caveats
・Data-retention policies vary by model. Many are zero-retention, but free models and some providers (e.g., OpenAI/Anthropic retain 30 days) may log or use your data. Verify before sending confidential data.
・Free models are "limited-time" offerings during feedback periods. Don't build permanent workflows on them.
・Some older-generation models are slated for deprecation by mid-2026. Monitor the deprecation dates of the models you use.
#OpenCode# #LLM#