Frequently Asked Questions

Everything you need to know about accessing Chinese AI models through NovAI's API gateway.

Getting Started

How can I access Chinese AI models like DeepSeek, Qwen, and GLM from outside China?

The easiest way is through an AI API gateway like NovAI. It provides a single OpenAI-compatible API endpoint at https://aiapi-pro.com/v1 that forwards your requests to Chinese AI model providers.

Here's how simple it is:

StepWhat you do
1. Sign upCreate a free account — get $2 credit instantly
2. Get API keyCopy from your dashboard
3. Call APIChange base_url and api_key in your code

No Chinese phone number, no VPN, no Alipay needed. Card (Visa/Mastercard/UnionPay) and USDT are both accepted.

Do I need to change my code to use Chinese AI models through NovAI?

No. NovAI provides a fully OpenAI-compatible API. If you're already using the OpenAI Python SDK, just change two values:

base_url = "https://aiapi-pro.com/v1"
api_key = "nvai-your-key-here"

Your existing code works immediately. It supports streaming, function calling, vision models, and all standard OpenAI features.

Is there a free trial for Chinese AI models?

Yes! Two ways to try for free:

  • GLM-4.6V-Flash — free to call; never-topped-up accounts run at 5 requests/minute with up to 10 free-model calls per day
  • $2.00 welcome credit — given to all new users upon email verification, enough to test premium models

No credit card required. Sign up free →

Pricing & Payment

How much does it cost to use Chinese AI models via API?

Chinese AI models are 60-80% cheaper than Western alternatives. Here's a comparison:

ModelInput / 1MOutput / 1Mvs GPT-4o
DeepSeek-V4-Flash$0.14$0.56~95% cheaper
Doubao-Seed-2.0-Lite$0.075$0.45~97% cheaper
GLM-4.6V-FlashFREEFREE—
Doubao-Seed-2.0-Pro$0.40$2.00~85% cheaper
Qwen3.7-Max$1.22$4.50~50% cheaper

See full pricing comparison →

What payment methods does NovAI accept?

Yes. NovAI accepts credit card (Visa / Mastercard / UnionPay) and USDT (TRC20) for API credit top-ups.

Most Chinese AI providers only accept Alipay or WeChat Pay, which require a Chinese bank account. NovAI bridges this gap — pay from anywhere in the world.

  • Minimum top-up: $1 by card (any amount); $5 for USDT / Alipay (preset amounts)
  • Credits never expire
  • First top-up bonus: $5+ = $1 extra credit
Does NovAI charge platform fees or markups?

No. NovAI charges zero platform markup. You pay exactly what the upstream provider charges.

Compare with competitors:

GatewayPlatform Fee
NovAI0%
OpenRouter5% markup
Together AIProvider markup varies
How much does AI video cost per minute?

Between $2.52 and $40.80 per minute of finished output, depending on model and resolution. The cheapest tier is Kling 2.6 at 720p ($0.084/s); the most expensive is Seedance 2.5 at 1080p ($0.68/s). Native audio adds about 50% on the same model.

Full per-minute table, every row sourced → · estimate your own batch →

How much does it cost to generate 1,000 images?

Roughly $30 to $204 at list price, depending on the model — from Wan 2.2 T2I Flash at $0.0215 per image to WAND Vega Image Pro at $0.146. Most of these models bill per image, not per megapixel, so 4K costs the same as 1K on many rows.

Run your own numbers →

Do I pay per attempt or per finished clip?

Per generated output. If a shot takes three takes, you pay for three. At a typical 1-in-2 hit rate, budget 2× the per-second price — a 300-clip month at 8 seconds is 4,800 generated seconds, not 2,400.

The calculator takes a retry rate →

How fast is the NovAI API?

Measured end-to-end at the gateway over 30 days across 63 models: the fastest models answer in about 324 ms at p50, the slowest long-output models take up to about 68 s. For short outputs the model dominates; for long ones generation time does.

Per-model p50 / p90 table, measured →

Is there a free tier, and what are the limits?

Yes. $2 of credit on signup, no card required, five free models, 10 free-model calls per day at 5 requests per minute, and up to 5 free video generations per day. Paid models are uncapped once you top up.

Current free-tier limits →

Are generated images watermarked?

Output from a never-topped-up account carries a NovAI watermark; once an account has any completed top-up, free-model output is clean. Separately, Chinese providers apply statutory AI-generated labels on some models — those come from the provider, not from us, and cannot be removed.

Watermark and labelling policy →

Performance & Models

Is NovAI faster than OpenRouter from Asia?

Yes, significantly. NovAI servers are in Hong Kong with direct peering to Chinese AI providers.

GatewayAvg Latency (from Asia)
NovAI (Hong Kong)75ms
DeepSeek Direct142ms
OpenRouter (US)207ms
OpenAI (US)217ms
Anthropic (US)242ms

For applications serving Asian users, NovAI reduces latency by 60-70%.

What Chinese AI models are available through NovAI?

111 models across text, image, video, audio and 3D, all through one API key:

  • ByteDance Doubao Seed 2.0 — Pro, Lite, Code (GPT-4-class, fast, code-specialist)
  • DeepSeek — V4 Pro, V4 Flash (reasoning, math)
  • Alibaba Qwen — Qwen3-Max, Qwen3.6-Max, Qwen3.7-Max, Qwen-Plus (multilingual, agentic)
  • Zhipu GLM — GLM-4.6V Flash, GLM-4.6V, GLM-5, GLM-5.1, GLM-5.2, GLM-5V-Turbo (vision, autonomous coding)
  • MiniMax — Text-01 (long-context generation)
What is the cheapest alternative to OpenRouter for Chinese AI models?

NovAI is cheaper for Chinese AI models because it charges 0% platform markup:

ModelNovAIOpenRouterSavings
DeepSeek-V4-Pro (input)$0.28$0.30+~7%
GLM-5.2 (input)$1.10$1.20~8%
GLM-5.2 (output)$3.80$4.10~7%

Plus lower latency from Hong Kong servers (75ms vs 207ms).

How do I compare DeepSeek vs Qwen vs GLM for my use case?

Each model family has different strengths:

  • DeepSeek-V4 — Best for reasoning, math, and scientific tasks
  • Qwen3.7-Max — Leads in multilingual, agentic, and tool-use tasks
  • GLM-5.1 — 8-hour autonomous coding, best for long-running development tasks
  • Doubao-Seed-2.0-Pro — GPT-4-class general performance at lower cost
  • GLM-4.6V-Flash — Free multimodal model for vision tasks

With NovAI's single API key, test all models side-by-side with the same prompts.

Trust & Safety

Why should I avoid unofficial AI API relays and resellers?

Unofficial relays (“中转站”) resell shared upstream API keys at a discount. The typical failure modes:

  • Shared quota & bans — hundreds of users on one key; when the provider detects abuse the key is banned and your integration dies overnight.
  • Silent downgrades — requests quietly routed to cheaper or older models (“降智”) while you pay flagship prices.
  • No billing trail — token counts and costs cannot be verified against the provider's official pricing.
  • Runaway risk — prepaid balances are lost when the relay disappears; there is no entity to hold accountable.
How is NovAI different from a relay?

Three structural differences:

  • Official direct endpoints — NovAI is a certified ecosystem partner of Volcano Engine (ByteDance) and Zhipu AI. Requests go straight to provider infrastructure; no reseller hops.
  • Dedicated quotas — your traffic runs on NovAI's own provider accounts, never on a shared black-market key pool.
  • Auditable billing — every call is logged in your dashboard with tokens, latency and cost, matched against provider list prices.

You can verify our partnership badges on the homepage, and compare every listed price with the provider's official rate card.

How can I check whether any AI API provider is trustworthy?

A quick due-diligence checklist:

  • Ask where the API keys come from — official partner accounts, or pooled third-party keys.
  • Check that prices match the upstream provider's official rate card (deep discounts usually mean shared keys).
  • Verify per-call usage logs with token counts and latency are available.
  • Start with a small top-up and confirm refund policy before committing volume.
  • Prefer providers with a registered business entity and public partnership certifications.

🚀 Ready to try Chinese AI models?

Sign up in 30 seconds. Get $2 free credit. No credit card needed.

Get Started Free →