DeepSeek-V4-Pro costs $2.57 per million input tokens and $7.71 per million output tokens through Adzuki AI. Cached input is $0.0857 per million, about 30× cheaper than fresh input. Billing is per token in USD against a prepaid balance with no subscription, and the model is called deepseek-v4-pro in the API. The context window is 1,000,000 tokens and a single response can run up to 384,000 tokens.
Every figure below is the price applied at request time.
| Model ID | deepseek-v4-pro |
|---|---|
| Vendor | DeepSeek |
| Input | $2.57 per 1M tokens |
| Cached input | $0.0857 per 1M tokens |
| Output | $7.71 per 1M tokens |
| Output vs input | 3.00× — output is the expensive side of every request |
| Context window | 1,000,000 tokens |
| Max output | 384,000 tokens per response |
| Rate limits | 60 requests/min · 100,000 tokens/min |
| Capabilities | text in / text out, implicit prompt caching, structured output, deep-thinking mode |
| Wire format | OpenAI Chat Completions — POST /chat/completions |
| Base URL | https://adzukiai.com/v1 |
| Billing | Per token, prepaid USD balance. No subscription, no monthly minimum. |
| Restrictions | Not for U.S. federal use. See the compliance note below. |
A worked example: a 10,000-token prompt where 8,000 tokens hit the cache, which produces 2,000 output tokens.
| Component | Tokens | Rate per 1M | Cost |
|---|---|---|---|
| Fresh input | 2,000 | $2.5714 | $0.0051428 |
| Cached input | 8,000 | $0.0857 | $0.0006856 |
| Output | 2,000 | $7.7143 | $0.0154286 |
| Total | 12,000 | — | $0.021257 |
The same request without cache hits would cost $0.0411426 — caching is what makes long, repeated system prompts affordable.
All seven models in the catalogue, sorted by input price, shown as a multiple of DeepSeek-V4-Pro. USD per million tokens.
| Model | Vendor | Input | Output | Input vs DeepSeek-V4-Pro |
|---|---|---|---|---|
| doubao-seed-2.0-lite | ByteDance Doubao | 0.17 | 1.03 | 0.07× |
| glm-5.3-flash | Zhipu AI | 0.23 | 0.80 | 0.09× |
| deepseek-v4.1-flash | DeepSeek | 0.57 | 2.29 | 0.22× |
| doubao-seed-2.1-pro | ByteDance Doubao | 1.71 | 8.57 | 0.67× |
| glm-5.3 | Zhipu AI | 2.29 | 8.00 | 0.89× |
| deepseek-v4-pro | DeepSeek | 2.57 | 7.71 | 1.00× |
| kimi-k3 | Moonshot AI | 5.71 | 28.57 | 2.22× |
No sales call, no application form.
Email and password. No phone number, no government ID.
Pay by card in USD. The balance is drawn down per token as you use it.
deepseek-v4-pro in any OpenAI SDK.
curl https://adzukiai.com/v1/chat/completions \ -H "Authorization: Bearer $ADZUKI_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "deepseek-v4-pro", "messages": [{"role": "user", "content": "Explain B-tree splits"}] }'
Two facts to check before you send production traffic.
Under the FY2026 National Defense Authorization Act (§1532 and §6604) and the U.S. Department of Homeland Security directive issued in October 2026, U.S. federal civilian agencies, the Department of Defense, the Intelligence Community and their contractors may not procure or use DeepSeek technology — including through third-party services such as this one, and regardless of whether the model is self-hosted. Several states, including New York, Virginia, Tennessee and Texas, have imposed comparable restrictions on government devices and networks.
If you are a U.S. federal agency, a federal contractor, or a state or local government entity, do not use the DeepSeek models offered here. Other models in the catalogue are not covered by this restriction. This notice is provided for compliance convenience and is not legal advice; you remain responsible for your own regulatory position.
Every model in this catalogue is operated by a provider whose inference infrastructure is located in mainland China. When you call a model, the content of your request is transmitted to that provider and processed there, and the response is returned to you. Account and billing data are stored in the United States and the European Union.
We do not use your prompts or outputs to train models. If your workload cannot leave a particular jurisdiction, contact support@adzukiai.com before sending production traffic.
Cost, caching and jurisdiction — the three things people ask first.
DeepSeek-V4-Pro costs $2.57 per million input tokens and $7.71 per million output tokens, with cached input at $0.0857 per million. Billing is per token against a prepaid USD balance; there is no subscription and no monthly minimum.
Yes. Cached input is $0.0857 per million tokens against $2.57 for fresh input, roughly 30× cheaper. Caching only applies when the provider serves a prompt prefix from its cache; the first request that populates the cache still pays the full input rate.
DeepSeek-V4-Pro accepts 1,000,000 tokens of context, up to 1,000,000 tokens in a single request, and can return up to 384,000 tokens in one response. The gateway allows 60 requests per minute and 100,000 tokens per minute per account.
No. Adzuki bills one flat per-token rate at every hour, so a request costs the same at 03:00 China time as at 15:00, and there is no surge tier to plan around. The upstream provider does charge half its list price between 22:00 and 08:00 China time, and that discount is not passed through — so work you can schedule overnight is the one case where a direct mainland account comes out cheaper.
No. DeepSeek-V4.1-Flash costs $0.57 per million input tokens and $2.29 per million output tokens, so DeepSeek-V4-Pro is 4.5× the input price and 3.4× the output price. In this catalogue the cheaper model is the budget tier and DeepSeek-V4-Pro the flagship tier.
The model is not restricted for general commercial use in the United States. It is restricted for U.S. federal use: federal civilian agencies, the Department of Defense, the Intelligence Community and their contractors may not procure or use DeepSeek technology, including through a third-party gateway such as this one.