OpenCode Zen — free tier setup

What you get free

A curated AI gateway (from the OpenCode team) with five free models: big-pickle, deepseek-v4-flash-free, mimo-v2.5-free, north-mini-code-free, nemotron-3-ultra-free. Per the catalog: "AI gateway with curated models. Free models may use data for improvement." Category: conditional_free — the "price" of the free models is your data: official docs say most Zen models are zero-retention, but free/trial models are the explicit exception and may retain prompts for model improvement. The other gotcha: these are trial-period models (some stealth/alias, like Big Pickle, with the underlying vendor undisclosed) — expect the free list to churn.

Friction check


Credit card	unknown — official onboarding includes an "add billing details" step, but it's unconfirmed whether it's enforced for free-only usage
Phone verification	unknown
Org/membership	none
Est. time to first key	~5–10m (4 steps)
Trains on your prompts	yes (free models may retain data for improvement)

Get a key

Go to opencode.ai/auth and sign in (the official docs onboarding starts here).
Add billing details to your account — this is step 2 of the official flow. Whether you can skip it for free models only is unconfirmed (unconfirmed — verify on first signup).
Copy your API key from the account/console page (exact page name unconfirmed — verify on first signup).
(Optional, TUI users) In OpenCode run /connect, select OpenCode Zen, and paste the key.

First request

One base URL, three endpoint shapes: /zen/v1/chat/completions and /zen/v1/responses (OpenAI-style), /zen/v1/messages (Anthropic-style). Expected success shape: a JSON chat.completion object with choices[0].message.content. Note: docs elsewhere show model ids prefixed as opencode/<model-id>; whether the bare catalog id or the prefixed form is required on the raw API is unconfirmed — try the bare id first (unconfirmed — verify on first request).

export OPENCODE_ZEN_API_KEY="YOUR_KEY"
curl https://opencode.ai/zen/v1/chat/completions \
  -H "Authorization: Bearer $OPENCODE_ZEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "big-pickle",
    "messages": [{"role": "user", "content": "ping"}],
    "max_tokens": 16
  }'

Limits that will bite you

No numeric rate limits are published for the free models — not on the official Zen docs, not in seed repos (catalog: units: []). Assume unadvertised throttles; back off on 429s.
The data policy is the real limit: free models may retain and train on prompts. Keep sensitive/production traffic off this provider.
Free model availability is trial-based and can disappear without notice; stealth models (big-pickle) can be swapped or removed.
Match model family to endpoint shape: OpenAI-format models on /chat/completions or /responses, Anthropic-format models on /messages.

Key rotation runbook

Second key: no multi-key documentation found on the Zen docs page (unconfirmed — verify in the console after signup).
Revoke: from the same account/console page where the key is shown (unconfirmed — verify on first signup).
ModelHubby integration: export OPENCODE_ZEN_API_KEY_2="..." — the key pool picks it up automatically; state machine cools down rate-limited keys per-key.

ModelHubby usage

pnpm modelhubby show opencode-zen
pnpm modelhubby probe --live --provider opencode-zen --apply