Prepaid credits or monthly plans · OpenAI-compatible New accounts get  500,000 quota ($1) free  on sign-up — up to 9M+ tokens · no card needed

Open-weight LLMs.
One OpenAI-compatible API.

Call DeepSeek, Qwen, GLM and Kimi from your existing code — same SDK, same tool-calling, no GPU to rent, no servers to run. Top up in USD with any international card: no Chinese phone number, Alipay or local bank required.

Not sure which model? Send "model": "auto" — we route each request to the best-fit model and you pay that model's normal rate, no routing fee. See how →

Bigger packs: $20 · $50 · credits never expire

No Chinese phone required Pay in USD by card Prepaid credits, never expire Failed requests not billed No prompt retention
terminal — bash
# Works with any OpenAI SDK — just change base_url & key
curl https://api.novatoken.dev/v1/chat/completions \
  -H "Authorization: Bearer sk-..." \
  -d '{"model": "auto", "messages": [{"role":"user","content":"Hello!"}]}'
How it works

Up and running in three steps

From zero to your first response in about a minute.

1

Create an account

Sign up with your email (one address, one account). Your account starts with 500,000 quota ($1) in free credit — no card needed to try it.

2

Change one line of code

Point your OpenAI SDK at https://api.novatoken.dev/v1 and paste your key. Cursor, Cline, LangChain, Open WebUI and most tools work out of the box.

3

Ship and monitor

Stream responses, call tools, route with auto. Track every request and its cost in your dashboard, and top up only when you need to.

Models & rates

Simple pricing, per token

Pay only for what you use. Prices in USD per 1M tokens. Don't want to pick? Send model: "auto" and we route each request to the best-fit model — you pay that model's own rate, no routing fee. How it works

Qwen2.5-7B-Instruct

Fast general chat · tool calling
$0.10 input / $0.30 output
  • 32K context window
  • No thinking delay — ~2s first token
  • Apache 2.0 licensed model
  • Great for chat and classification
Buy $5 credits

GLM-5.3-Flash

Fast reasoning · strong Chinese + English
$0.15 input / $0.50 output
  • 1M token context window
  • Quick reasoning turns (~4s)
  • Solid on bilingual tasks
  • Agents and summarisation
Buy $20 credits

Kimi-K2.7-Code

Coding specialist · agentic workflows
$1.20 input / $5.00 output
  • 262K context window
  • Tuned for code and tool use
  • Long repository style inputs
  • Best for Cursor / Cline style tools
Buy $20 credits

Need a specific model sooner? Email [email protected]

Pricing

Top up your balance

Prepaid credits that never expire, or a monthly plan with credits auto-refilled — both work with every model we serve. Card payments are processed by Creem, our merchant of record.

$5 credits

Try it out · side projects
$5 one-time
  • ≈ 50M input tokens on Qwen2.5-7B
  • ≈ 16M input tokens on DeepSeek-V4-Flash
  • No expiry, no monthly fee
Buy $5

$50 credits

Teams · high volume
$50 one-time
  • ≈ 500M input tokens on Qwen2.5-7B
  • ≈ 166M input tokens on DeepSeek-V4-Flash
  • Best value per token
Buy $50

Checkout is locked to the account you are logged into, so credits are always added to the right balance — usually within a minute.

Or go monthly

One flat price, credits auto-refill on your billing date, cancel anytime.
Launch promo: first month 50% off with code PRODUCTHUNT at checkout.

Max

$29.90 / month
  • 4,000 credits every month (≈ $40 of usage)
  • Auto-refills on your billing date
  • Priority routing capacity
  • Email support
  • Cancel anytime, keep unused balance
Choose Max
Why NovaToken

Why developers switch

The boring things, done right.

Drop-in compatible

Change one line — base_url and API key. Chat completions, SSE streaming, tool calling and JSON mode all work, so the rest of your stack stays untouched.

Transparent billing

Real-time usage dashboard with per-request cost breakdown. Failed requests are never billed, there is no minimum spend, and you top up only when you want.

Resilient routing

Requests run on official upstream channels with automatic fallback to secondary providers, so a single provider outage does not take your app down.

Your prompts stay yours

No prompt retention and no training on your data. We keep timestamps and token counts for billing only — see our Privacy Policy for details.

Start shipping today.

Create an account, grab a key, make your first call in under a minute — $1 in free credit is already on your balance. Top up $5 whenever you're ready.

Buy $5 of credits Create free account
FAQ

Frequently asked questions

Which model should I pick?

If you don't want to think about it, use auto. It looks at each request and sends it to the best-fit model — short chat to Qwen2.5-7B, everyday answers to GLM-5.3, hard reasoning and code to DeepSeek-V4 — and you are billed at that model's normal rate, exactly as if you had chosen it yourself (no routing fee). Your usage log records which model actually ran. Pick a specific model only when you know you need it, for example moonshotai/Kimi-K2.7-Code for heavy coding agents. Full guide

Is it really compatible with the OpenAI SDK?

Yes — chat completions, streaming (SSE), tool/function calling and JSON mode are supported. Set base_url to https://api.novatoken.dev/v1 and use your NovaToken key.

Do I need a Chinese phone number or Alipay?

No. You register with an email address and pay in USD with any major credit or debit card, processed by Creem, our merchant of record. No Chinese phone number, bank account or payment app is involved anywhere.

Do repeated prompts cost less?

Yes, on DeepSeek-V4-Flash. When your request reuses context that is still in the provider's cache — common with long system prompts, agents and coding assistants — those input tokens are billed at $0.006 per million instead of $0.30, a 50x saving. You do not have to do anything to enable it; the discount is applied automatically whenever the upstream reports a cache hit. The other three models do not offer prompt caching, so every input token is billed at the list price.

Are you cheaper than the official APIs?

No — we are a reseller, not a subsidy. Our per-token prices sit close to the Chinese list prices, and the markup covers card processing, payment-provider fees and support. If you already have a Chinese phone number, Alipay and a Chinese bank account, going direct will usually cost you a little less. What you are paying for here is access: no Chinese phone number, no Alipay or WeChat Pay, no Chinese invoicing, one USD card checkout and one endpoint for four model families.

What do I get for free?

Every new account starts with 500,000 quota ($1.00) in free credit — enough for roughly 2M DeepSeek tokens, 6M Qwen tokens or 4M GLM tokens. No card required and the credit never expires. Registration is verified by email, and one email address maps to one account (plus-aliases and disposable domains are rejected).

Do you offer monthly plans?

Yes. Pro is $9.90/month with 1,000 credits (≈ $10 of usage) auto-refilled on your billing date, and Max is $29.90/month with 4,000 credits (≈ $40 of usage) plus priority routing capacity and email support. Both include every model and auto routing, and you can cancel anytime — unused balance stays on your account. Right now the first month is 50% off with code PRODUCTHUNT at checkout.

Do you store my prompts?

No. We record timestamps and token counts for billing, and that metadata is purged on a rolling basis. Request bodies are not retained and your data is never used for training.

Some models pause a few seconds before answering — why?

DeepSeek-V4, GLM-5.3 and Kimi-K2.7 are reasoning models: they think before answering, and thinking tokens are billed as output. With stream: true you still get the first byte fast. If you want zero thinking delay, use Qwen/Qwen2.5-7B-Instruct.

Which payment methods do you accept?

Major credit and debit cards worldwide, processed by Creem (merchant of record) and charged in USD. Receipts are issued by Creem; taxes are handled by them as well.

What happens if a request fails or the service is down?

Failed requests are not billed — you only pay for completed responses. Our routing uses official upstream channels with automatic fallback to secondary providers, so a single provider outage should not interrupt your app. Credit refunds follow the Refund Policy.