DeepSeek V4 API With USDT or Card: Use the OpenAI SDK

DeepSeek V4 Flash and Pro are strong open models for agents and coding, and paying for them can be awkward when an international checkout declines your card. cloudgpu.app runs an OpenAI-compatible gateway in front of DeepSeek, GLM, Kimi, MiniMax and a few image and speech models, paid from one prepaid balance that you top up with USDT or a card. This post covers setup, topping up with USDT, and what the gateway does and does not do.

What it costs

The per-token rate for every model is on the API page. Rates change, so this post does not repeat them. You pay per token from a prepaid balance, with no subscription. The balance does not expire.

Set it up

1. Sign up at cloudgpu.app with an email address or a Google account.
2. In the API console, create a key. It starts with cgw-sk- and is shown only once, so copy it when it appears.
3. Point your client at the base URL https://cloudgpu.app/v1.

curl:

curl https://cloudgpu.app/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cgw-sk-..." \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Summarise the CAP theorem in two sentences."}]
  }'

Python, with the OpenAI SDK and streaming:

from openai import OpenAI

client = OpenAI(
    base_url="https://cloudgpu.app/v1",
    api_key="cgw-sk-...",
)

resp = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Write a haiku about GPUs."}],
    stream=True,
)
for chunk in resp:
    print(chunk.choices[0].delta.content or "", end="")

Cursor:

• Settings → Models → OpenAI API Key: paste your cgw-sk-... key.
• Turn on Override OpenAI Base URL and enter https://cloudgpu.app/v1.
• Add deepseek-v4-flash as a custom model and make sure it is enabled.

Cline, Continue or OpenClaw: choose the OpenAI-compatible provider, use the same base URL and key, and enter the model ID as listed on the API page.

Topping up with USDT

1. On the billing page, choose USDT (TRC-20). The page shows a TRON address and the exact amount to send.
2. In your exchange or wallet, send USDT for that amount and select the TRON (TRC-20) network.
3. Wait for the transfer to confirm. The balance on the billing page updates once it is credited, in US dollars at face value.

Send only USDT, and only on TRON. USDT on any other network, and any other coin, is not credited. Send the amount the page shows, to the cent: that is how the transfer is matched to your account.

If you would rather not use crypto, the card checkout also offers Alipay, Google Pay and Apple Pay, and business customers can pay by bank transfer.

What the gateway does for you

• One endpoint. Change base_url once and keep using the OpenAI SDK.
• One balance for DeepSeek, GLM, Kimi and MiniMax, instead of a separate account and payment method for each vendor.
• The same balance pays for GPUs. If you also rent GPUs on cloudgpu.app, prototype against the hosted API and move heavy workloads to a rented GPU running vLLM, on one ledger.

What it does not do

• It is not free: you pay per token from a prepaid balance.
• It does not make you anonymous. Sign-up needs a real email address, and abuse such as spam or anything illegal gets a key revoked.
• It is not a custom model host. To run your own model, rent a GPU on the deploy page and run vLLM or Ollama on it.
• There is no automatic failover. If a model’s upstream is down, that model is down here too.

Which model for what

• deepseek-v4-flash: the default for agent and chat traffic.
• deepseek-v4-pro: when Flash is not enough, for hard reasoning and long multi-step tool use.
• glm-5.3-flash: classification, extraction and other high-volume, low-stakes calls.
• kimi-k3: long documents, and a second opinion on hard problems.
• kimi-k2.7-code: coding agents.
• minimax-m3: long-context work.

The full model list is on the API page.

Data and jurisdiction

The gateway forwards each request to the infrastructure serving that model. Open-weights models are served from each vendor’s published weights; for deepseek-v4-flash that is the April open-weights checkpoint, not the newer V4.1 Flash. Do not send data that your contracts or laws require to stay in a specific jurisdiction. The gateway does not train on your prompts and does not store request bodies; token counts, the model name and error codes are kept for billing and error diagnosis.

Start

Create an API key, top up a small amount, and run your real workload for a day before you add more.

Frequently asked questions

Do I need a phone number or a credit card?

No. Sign up with an email address or a Google account, then top up with USDT on the TRC-20 network or through the card checkout.

Does it work with the OpenAI SDK?

Yes. Set base_url to https://cloudgpu.app/v1 and use a model ID such as deepseek-v4-flash.

Where are the current rates?

On the API page. You pay per token from a prepaid balance, and rates can change.

Which USDT network should I use?

TRON (TRC-20) only. USDT on any other network, and any other coin, is not credited.