How to Use DeepSeek API Without a Chinese Phone Number

If you have tried to sign up for the DeepSeek API from outside mainland China, you already know where it stops: the registration flow expects a Chinese mobile number. On top of that, topping up the account generally requires a domestic payment method such as Alipay or WeChat Pay tied to a Chinese bank account.

The model itself has no such restriction. Nothing about DeepSeek's weights, context window, or reasoning quality is region-locked — the friction is entirely in the account layer.

This guide shows the shortest path around it. You will get a working DeepSeek API key with an email address or a Google/GitHub login, pay with an international card, and make your first call in roughly three minutes, using the OpenAI SDK you probably already have installed.

---

The short answer

Question	Official DeepSeek platform	TokenPAPA
Phone number required	Mainland China (+86) mobile	None
Signup method	Phone + SMS code	Email, or Google / GitHub OAuth
Payment	Domestic methods (Alipay / WeChat Pay)	International cards, Apple Pay, Google Pay
Minimum top-up	Varies by channel	$10
API format	OpenAI-compatible	OpenAI-compatible
Base URL	https://api.deepseek.com	https://tokenpapa.ai/v1
SDK changes	—	Change base_url, api_key, model

Same models, same request/response shape, same client library. Only the account layer changes.

---

Why the official route is hard from overseas

Three separate gates line up in sequence, and each one is a hard stop on its own:

1. Registration: the signup form is built around a Chinese mobile number and an SMS verification code. That number is not something most overseas developers have.
2. Payment: even with an account, funding it typically goes through domestic payment rails. A US, EU, or Southeast Asian card often does not work.
3. Support and billing: invoices, receipts, and account recovery assume a domestic identity and bank.

There is no switch to flip. The DeepSeek platform is designed first for the domestic market, and overseas access is a secondary use case there.

The alternative is to go through a gateway that already has the upstream relationship — which is exactly what TokenPAPA does. You get a normal account on the international side; TokenPAPA holds the China-side relationship and routes your requests.

---

Three routes, honestly compared

Route	Setup effort	Recurring maintenance	Best for
Official DeepSeek platform	Blocked for most overseas devs	—	Developers with a Chinese number and bank account
Self-hosting the open weights	Days of setup, GPU rental	Model updates, scaling, ops	Teams with privacy requirements and real infra budget
TokenPAPA gateway	About 3 minutes	None — one key, managed routing	Anyone who wants DeepSeek plus 60+ other models today

Self-hosting is a legitimate answer for a narrow set of teams, but it is not a shortcut: you are now responsible for GPUs, quantization trade-offs, throughput, and every future model upgrade. For most developers, the gateway is the pragmatic route.

---

Step 1 — Create an account

Go to tokenpapa.ai and register with an email address, or use Google or GitHub one-click login. First-time OAuth login creates the account automatically.

No phone number, no SMS code, no Chinese identity document.

Step 2 — Create an API key

Open the console and generate an API key. Copy it once and store it somewhere safe — treat it exactly like an OpenAI key. Do not commit it to a repository.

Step 3 — Send your first request

Install or reuse the OpenAI SDK and point it at TokenPAPA:

from openai import OpenAI

client = OpenAI(
    api_key="your-tokenpapa-key",
    base_url="https://tokenpapa.ai/v1"
)

resp = client.chat.completions.create(
    model="deepseek-v4-flash",     # DeepSeek V4 Flash
    messages=[
        {"role": "user", "content": "Summarize why an LLM API gateway is useful in three bullets."}
    ],
    max_tokens=300                 # always cap output — output tokens cost more than input
)

print(resp.choices[0].message.content)

That is the entire migration. If your code already talks to OpenAI, you are changing three values and nothing else — base_url, api_key, and model.

---

What you can switch to afterwards

The practical reason to route DeepSeek through a gateway is that the same key also reaches the rest of the field. Changing models is a one-line edit:

Model ID	Good at	Input / 1M	Output / 1M
deepseek-v4-flash	Cost-effective general work, agentic coding	$0.14	$0.42
deepseek-flash (V4.1 Flash)	Newer DeepSeek Flash line, cache-friendly	$0.30	$1.20
deepseek-v4-pro	Heavier reasoning and long-form work	$0.28	$0.84
qwen3.7-plus	Coding, structured output	$0.20	$0.60
kimi-k3	Long-context document work (256K)	$0.50	$2.00
gpt-5.6-luna	Budget OpenAI tier, 1M context	$0.27	$2.70
claude-sonnet-4-6	Careful writing, refactors, review	$3.00	$15.00

Rates move. Treat this table as a starting point and confirm current numbers on the pricing page before you plan a budget.

One detail worth knowing early: cached input is dramatically cheaper than fresh input on DeepSeek. deepseek-flash lists cached input at $0.006 per 1M tokens. If your app reuses a stable system prompt, keep that prefix byte-identical between requests and the input side of your bill collapses.

---

Common errors and what they actually mean

Error	Meaning	Fix
401 Invalid token	Key is wrong, revoked, or has a whitespace typo	Re-copy the key; check for a trailing newline in your env var
402 Insufficient balance	Account balance exhausted	Top up in the console (minimum $10)
404 model not found	Model ID does not exist on the platform	Use an ID from the live model list, e.g. deepseek-v4-flash
429 Too many requests	You hit a rate limit	Add exponential backoff with a jitter, and retry
400 context length exceeded	Prompt plus max_tokens exceeds the window	Trim history, or move to a longer-context model

The 404 case trips people up most often, because marketing names and API model IDs do not always match. For example, the model marketed as DeepSeek V4.1 Flash is called with the ID deepseek-flash — no version number, no dot. Always confirm the exact ID before shipping code.

---

FAQ

Q: Can I use the DeepSeek API without a Chinese phone number?
A: Yes. The official platform requires a mainland China mobile number and a domestic payment method. TokenPAPA removes both: sign up with an email address or Google/GitHub login, then top up with an international card. There is no phone verification step.

Q: Is the DeepSeek API on TokenPAPA compatible with the OpenAI SDK?
A: Yes. The endpoint at https://tokenpapa.ai/v1 is OpenAI-compatible. Keep your existing client, change base_url and api_key, and set model to a DeepSeek ID such as deepseek-v4-flash.

Q: How much does the DeepSeek API cost on TokenPAPA?
A: DeepSeek V4 Flash is listed at $0.14 per 1M input and $0.42 per 1M output tokens. DeepSeek V4.1 Flash (deepseek-flash) is listed at $0.30 per 1M input, $1.20 per 1M output, and $0.006 per 1M cached input. Confirm current rates at tokenpapa.ai/pricing before budgeting.

Q: How do I pay for a DeepSeek API key from outside China?
A: TokenPAPA accepts international channels: Stripe (bank cards, Apple Pay, WeChat Pay, Alipay) and Waffo Pancake (Google Pay, Apple Pay, bank cards). The minimum top-up is $10.

---

Get Started

1. Sign up at tokenpapa.ai — email, Google, or GitHub. No Chinese phone number.
2. Create an API key in the console.
3. Point your client at https://tokenpapa.ai/v1 and call deepseek-v4-flash.

from openai import OpenAI

client = OpenAI(api_key="your-tokenpapa-key", base_url="https://tokenpapa.ai/v1")

print(client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Hello from outside China."}],
    max_tokens=100
).choices[0].message.content)

Three minutes of setup, one key, and DeepSeek V4 Flash plus the rest of the field — including Qwen, Kimi, GLM, GPT, Claude, and Gemini — is reachable from any country with the SDK you already use.

Originally published at canonical.