Free cloud APIs

MiniTavern is BYOK (bring your own key): it does not include cloud credits. You register with a vendor or inference platform and paste your own API Key.
This page summarizes common free / trial quotas and how to fill them into the App. Quotas and policies change — always trust what each vendor’s console shows in real time.

Security note: an API Key is like a password. Store it only in the App’s on-device secure storage. Do not paste it into group chats, public screenshots, or public repos.

Fastest path

Your situation	Suggestion	App platform
Mainland China, want to chat ASAP	Zhipu Flash free models or SiliconFlow free models	Custom (OpenAI)
One Key for text + image (multimodal)	Agnes AI	Custom (OpenAI) / Images
One Key for GPT / Gemini *-free + image	AIHubMix	Custom (OpenAI) / Images
Coding-oriented gateway, limited Free text	OpenCode Zen	Custom (OpenAI)
Official Chinese LLM trial	Qwen / Alibaba Cloud Bailian	Qwen
Reasoning / code-oriented	DeepSeek	DeepSeek
Overseas account, multimodal text	Google Gemini	Gemini(Google)
One Key for many free open-source models	OpenRouter :free	OpenRouter
Not sure which free vendors exist	FreeLLM directory (awesome-freellm-apis)	See directory
Ultra-fast open-source inference	Groq	Custom (OpenAI)
Prefer no cloud	Run local models	Custom (OpenAI)

General Key steps: Model link. Local deploy: Run local models · 30-minute quickstart.

How to fill in MiniTavern

Path: Settings → Models → Model link → New

1. Pick the correct model platform (see “App platform” below).
2. Paste the API Key; for custom platforms also fill Base URL.
3. Select or type the model ID (must match vendor docs; case-sensitive).
4. Tap Test → on success, set as Default.

Vendor	App platform	Base URL (common)	Notes
Gemini	Gemini(Google)	Preset	Gemini protocol — do not fill as OpenAI
OpenRouter	OpenRouter	Preset	Free model IDs often end with :free
Qwen / Bailian	Qwen	Preset (intl switchable)	Use a general API Key, not a Coding Plan–only Key
DeepSeek	DeepSeek	Preset
Zhipu / SiliconFlow / Agnes / AIHubMix / Zen / Groq / Mistral, etc.	Custom (OpenAI)	See each section	OpenAI-compatible gateways

---

China-focused (prefer mainland direct access)

Zhipu AI (GLM)

Best for: long-term free chat (Flash series); friendly for Chinese roleplay.

Item	Details
Site / Key	open.bigmodel.cn → Console → API Keys
Free tier	Permanent free models such as GLM-*-Flash; new users often get extra trial credit (see console)
Card required	Usually no
Base URL	https://open.bigmodel.cn/api/paas/v4
Model examples	glm-4.5-flash, glm-4.7-flash (names follow the model plaza)
App	Custom (OpenAI)

Steps: register → create API Key → in App choose Custom (OpenAI) → Base URL + Key → Flash-series model ID → test.

SiliconFlow

Best for: one Key across many open-source chat models; some image models also have free tiers.

Item	Details
Site / Key	cloud.siliconflow.cn → Account → API keys
Free tier	Free models are not billed (name without Pro/ prefix); complete real-name verification for full free catalog
Rate limits	Fixed RPM/TPM per model; see official Rate Limits
Base URL	https://api.siliconflow.cn/v1
Model examples	In the plaza, filter “Free”, e.g. Qwen/Qwen3-8B (list changes)
App	Custom (OpenAI)

Note: the same name may have a free version and a paid Pro/... version — wrong ID bills you. Free-model usage on the bill should be 0.

Qwen / Alibaba Cloud Bailian

Best for: official Qwen series; MiniTavern has a Qwen preset.

Item	Details
Activate	Bailian console (China North 2 / Beijing) — accept terms to auto-activate
Free tier	New-user free quota: most models ~1M tokens / model, ~90 days; quotas do not share across models
Key	Bailian → API-Key; use a general Key (Token Plan / Coding Plan–only Keys do not consume free quota)
App	Qwen (China preset; international site switchable)
Avoid charges	Enable “Stop when free quota is used up” so pay-as-you-go does not start after quota ends

Official docs: new-user free quota. Calls after quota exhaustion or expiry incur charges.

Image gen (Wanxiang / Qwen-Image, etc.) may use new-user quota — in Settings → Image gen → Image gen link pick Qwen; whether it is free depends on that model’s quota bar in the console.

DeepSeek

Best for: reasoning, long text, high value; cheap after credits, but not permanently free.

Item	Details
Site / Key	platform.deepseek.com → API Keys
Free tier	New users usually get signup credit / trial balance (amount and rules change; see console)
App	DeepSeek
Model examples	deepseek-v4-flash, deepseek-v4-pro (see official docs)

After credits run out, usage is billed; requests fail with zero balance.

ModelScope API-Inference

Best for: Alibaba Cloud real-name already done; try community open-source inference.

Item	Details
Token	modelscope.cn → Access token
Free tier	Registered users get daily API-Inference quotas (totals and per-model caps: official limits)
Base URL	https://api-inference.modelscope.cn/v1
App	Custom (OpenAI)

---

International / aggregators

Agnes AI

Best for: one free Key for text chat and image gen (both Flash free tier). Video is on the site; the App has no built-in video entry yet.

Full intro, model tables, and Key steps: Agnes AI.

Item	Details
Site / Key	agnes-ai.com · platform.agnes-ai.com
Free tier	Free Access: Flash text / image callable for free (RPM limits)
Base URL	https://apihub.agnes-ai.com/v1
Text example	agnes-2.5-flash
Image example	agnes-image-2.1-flash (image link → Custom (OpenAI Images))
App	Custom (OpenAI) / Custom (OpenAI Images)

OpenCode Zen

Best for: Free text models on the OpenCode curated gateway (e.g. mimo-v2.5-free). Most flagships are metered — watch auto-recharge.

Full guide: OpenCode Zen.

Item	Details
Site / Key	opencode.ai/auth · docs
Free tier	Models marked Free on the pricing table (within caps); others are metered
Base URL	https://opencode.ai/zen/v1
Text examples	mimo-v2.5-free, hy3-free, big-pickle
App	Custom (OpenAI) (prefer chat/completions-compatible IDs)

AIHubMix

Best for: one Key for many platform-subsidized *-free models (GPT / Gemini / GLM, etc.) and GPT-Image-2-free image gen; usually no card required.

Full guide: AIHubMix.

Item	Details
Site / Key	aihubmix.com · free models post
Free tier	Model IDs ending in -free; per-model RPM / daily caps
Base URL	https://aihubmix.com/v1
Text examples	gpt-4o-free, gpt-5.5-free, coding-glm-5.1-free
Image example	gpt-image-2-free (Custom (OpenAI Images))
App	Custom (OpenAI) / Custom (OpenAI Images)

Google Gemini

Best for: multimodal text, long context; text free tier is relatively generous.

Item	Details
Key	aistudio.google.com/app/apikey
Free tier	Some Gemini Flash / Lite text models have a Free tier (RPM / RPD as shown in AI Studio)
Card required	Text free tier usually no; latest image models mostly have no Free tier
Region	EU / UK / Switzerland may lack free tier
Privacy	Free-tier prompts may be used to improve products (see Google terms)
App	Gemini(Google)

For chat, pick a text model (e.g. gemini-flash-latest). Do not pick IDs with tts or image: Test sends a text reply request; TTS only outputs AUDIO and errors with response modalities (TEXT) is not supported. Configure voice/image under those features.

If Test/send returns HTTP 404 with no longer available to new users, that model ID is retired or closed to new users (e.g. gemini-2.5-flash-lite). Refresh the model list and pick a live Flash / Lite (e.g. gemini-flash-latest). You do not need the Interactions API.

If send fails with HTTP 429 and generate_content_free_tier_requests (often limit: 20), free-tier RPM is exhausted — wait for Please retry in … then retry. See ai.dev/rate-limit; raise limits via billing in AI Studio. This is not “no money” or a monthly project spend cap.

Image gen: Gemini App web and Developer API quotas differ; API image pricing pages often mark Free tier unavailable. Chat on Gemini + local A1111/ComfyUI for images is a common split — see Image model deploy.

OpenRouter

Best for: one Key across a large open-source / third-party free pool; IDs often end with :free. Full lists, tables, and troubleshooting: OpenRouter.

Item	Details
Key	openrouter.ai/keys
Free tier	:free / Free collections; default ~20 RPM, limited daily requests; topping up can raise free daily caps (official rules)
App	OpenRouter
Model examples	openrouter/free, meta-llama/llama-3.3-70b-instruct:free, openai/gpt-oss-20b:free
Collections	Free Models · summary openairouter.net/free-models

Note: free upstreams may log prompts for training. On HTTP 429: Provider returned error, try another :free, wait, or BYOK.

Groq

Best for: very low latency open-source inference.

Item	Details
Key	console.groq.com/keys
Free tier	Permanent Free tier (RPM / RPD / TPM per model; changes over time)
Card required	Usually no
Base URL	https://api.groq.com/openai/v1
Model examples	llama-3.1-8b-instant, llama-3.3-70b-versatile, openai/gpt-oss-20b
App	Custom (OpenAI)

Mistral AI

Item	Details
Key	console.mistral.ai/api-keys
Free tier	Experiment / trial plans (monthly token scale per console); prompts may be used to improve products
Base URL	https://api.mistral.ai/v1
App	Custom (OpenAI)

Cerebras / SambaNova and other fast inference

Platform	Key entry	Base URL	Notes
Cerebras	cloud.cerebras.ai	https://api.cerebras.ai/v1	Free tier often has daily token caps; may require a payment method
SambaNova	cloud.sambanova.ai	https://api.sambanova.ai/v1	Permanent free tier + occasional credits; strict rate limits

App: Custom (OpenAI) for both; model names follow each console.

GitHub Models

Item	Details
Entry	GitHub Models
Auth	GitHub Personal Access Token
Base URL	https://models.github.ai/inference
App	Custom (OpenAI)
Limits	Per-model RPM / RPD tiers — light trials

NVIDIA NIM

Item	Details
Entry	build.nvidia.com
Base URL	https://integrate.api.nvidia.com/v1
App	Custom (OpenAI) (or Nvidia preset if restored in App later)

---

Trial credits (not permanently free)

These are mostly time-limited credits — fine for evaluation, not as your only long-term backend:

Vendor	Typical situation	App
OpenAI	Occasional console trial credit; main API has no stable permanent free tier	ChatGPT(OpenAI)
Anthropic Claude	Console starter credits, burn quickly	Claude(Anthropic)
DeepSeek	Signup credit above	DeepSeek
Alibaba Cloud Bailian	New-user token quota (~90 days)	Qwen

When credits run out, reduce auto-charge risk (delete Key, enable stop-on-quota-end, or switch to free models / local).

---

Free image generation

Path	Cost	Notes
Local A1111 / ComfyUI	Power + GPU	Unlimited images; phone → PC LAN
SiliconFlow free image models	Free models = ¥0	Image link → Custom (OpenAI Images), same SiliconFlow Base URL …/v1, pick a free image ID
Agnes AI image Flash	Free Access	Image link → Custom (OpenAI Images), Base URL https://apihub.agnes-ai.com/v1, e.g. agnes-image-2.1-flash
AIHubMix	*-free image models	Image link → Custom (OpenAI Images), Base URL https://aihubmix.com/v1, e.g. gpt-image-2-free
Wanxiang / Qwen image	Depends on Bailian free quota for that model	Settings → Image gen → Image gen link → Qwen
OpenAI / most Gemini image APIs	Usually paid	Do not treat as “free image” without a stable Free tier

Chat and image links can differ: e.g. chat on Zhipu Flash, images on home ComfyUI.

---

Tips and troubleshooting

1. Test before chatting: on failure tap “View details” and check Error codes.
2. 429 / rate limits: lighter model, shorter memory, off-peak, or a second free platform.
3. Unexpected charges: check for Pro/ paid models, Bailian without stop-on-quota-end, or exhausted credits.
4. Privacy: free tiers often allow vendors to use logs to improve models; for sensitive content prefer local models.
5. Quota changes: numbers here were common at research time; trust the vendor site. Broader index: FreeLLM directory (awesome-freellm-apis · freellm.net, unofficial, updated daily).

Related

• Model link — how to fill fields in the App
• Getting started
• FreeLLM directory — more free vendors
• Run local models — local language / image services
• Image model deploy — A1111 / ComfyUI