OpenCode Zen
An AI gateway from the OpenCode team: curated, tested coding-oriented models, billed per request; plus several Free (time/cap-limited) text models. Usable outside OpenCode clients (including MiniTavern). Official docs: Zen (简体).
Zen is not a permanently free-only platform: most flagships are metered and often need billing / top-up; the “free models” below are for trials within caps. Trust the official site for policy.
Sites and entry points
| Resource | URL |
|---|---|
| Zen intro | opencode.ai/zen |
| Sign-in / Key | opencode.ai/auth |
| Docs (简体) | opencode.ai/docs/zh-cn/zen |
| Model metadata | https://opencode.ai/zen/v1/models |
| API Base URL (OpenAI-compatible) | https://opencode.ai/zen/v1 |
How to get free access
- Open opencode.ai/auth (GitHub / Google) and sign in.
- Create an API Key per the console. Official flows often ask for billing info; if you only want Free, still check that auto-recharge is off to avoid surprise charges.
- In MiniTavern pick Custom (OpenAI), Base URL
https://opencode.ai/zen/v1, model = a Free ID below.
| Item | Details |
|---|---|
| Free path | Some models priced Free (within OpenCode caps); used to gather feedback — may change or retire anytime |
| Paid path | Pay-as-you-go; below $5 balance, default auto-recharge $20 (change or disable in console) |
| Card required | Official “how it works” says add billing info; follow the live console |
| Privacy | Free tiers (e.g. Big Pickle, MiMo Free, Hy3, Nemotron Free, Muse Contributor Free) may use logs to improve models; Nemotron is an NVIDIA trial endpoint — do not send secrets |
Full pricing and deprecations: Zen · Pricing.
Model list
Model IDs are case-sensitive. Zen routes vendors differently: /chat/completions, /responses, /messages, or Gemini-specific paths. MiniTavern Custom (OpenAI) uses Chat Completions — prefer the OpenAI-compatible rows below.
Full table: Zen · Endpoints; or GET https://opencode.ai/zen/v1/models.
Text · Free (try these first)
All use …/v1/chat/completions except Muse Contributor Free (Responses).
| Model ID | Notes |
|---|---|
mimo-v2.5-free | Xiaomi MiMo V2.5 free |
hy3-free | Hy3 Free |
big-pickle | Stealth model, free |
nemotron-3-ultra-free | NVIDIA Nemotron trial — no secrets |
nemotron-3.5-lightning-free | Same |
muse-spark-1.2-contributor-free | Contributor free; uses /responses — MiniTavern Test may not apply; allows training Meta models on prompts |
Text · Paid but OpenAI-compatible (optional)
Same Base URL after you top up:
| Model ID | Notes |
|---|---|
deepseek-v4-flash / deepseek-v4-pro | DeepSeek (Peak / Off-Peak pricing) |
glm-5.1 / glm-5.2 / glm-5 | Zhipu GLM |
kimi-k2.5 / kimi-k2.6 / kimi-k2.7-code / kimi-k3 | Kimi |
minimax-m2.5 / minimax-m2.7 / minimax-m3 | MiniMax |
Other protocols (use carefully in MiniTavern)
| Type | Example IDs | Endpoint type |
|---|---|---|
| GPT series | gpt-5.5, gpt-5.4-nano, etc. | /zen/v1/responses |
| Claude series | claude-sonnet-4-6, claude-haiku-4-5, etc. | /zen/v1/messages |
| Gemini series | gemini-3-flash, etc. | /zen/v1/models/<id> |
These are not the same path as MiniTavern’s default OpenAI Chat Completions. For roleplay chat, prefer Free or OpenAI-compatible models above. For Claude, use App Claude(Anthropic) with an official Key, or another unified OpenAI gateway such as AIHubMix.
Images / video
Zen’s current catalog is mostly language models — no separate free image / video tier. For image gen see Agnes AI, AIHubMix, or Free image gen.
How to fill in MiniTavern
General steps: Model link.
Path: Settings → Models → Model link → New
- Platform: Custom (OpenAI)
- Base URL:
https://opencode.ai/zen/v1 - Paste API Key
- Model ID:
mimo-v2.5-free(or another Free / chat-compatible ID) - Test → on success, set as default
If Test fails: confirm you did not pick GPT Responses / Claude Messages–only IDs; disable auto-recharge before trying Free; see Error codes.
Related
- Free cloud overview
- AIHubMix — one Key for many
*-freeclosed/open models + image gen - Agnes AI
- Model link