For developers in China, paying for OpenAI, Anthropic Claude, or Google Gemini APIs is usually harder than using the APIs themselves. All three providers require a foreign credit card; none of them accept WeChat Pay or Alipay. If you are a developer based in mainland China, you probably do not have that card, or you have one but it comes with friction — FX conversion fees, limited annual quota, and no invoice that your accounting department will accept.
This post surveys the real options developers in China use to pay for LLM APIs, compares their risks, and walks through how to use WeChat Pay or Alipay to top up major models through Router One — covering individual developer and team/company scenarios.
Three Payment Frictions for Chinese Developers
One: foreign credit card requirement. OpenAI requires you to bind a credit card issued in the US or specific supported countries. Anthropic has similar restrictions. A Visa or Mastercard issued by a Chinese bank may or may not work — many banks issue international-brand cards but their BIN ranges are recognized as Chinese and rejected. The set of accepted cards keeps shrinking.
Two: FX rate and quota. Even when the card works, payments are in USD, charged at the central bank's daily rate plus a 1–2% conversion fee from the issuer. Individuals in China have a $50,000 annual FX quota that a serious team running production agents can blow through in a month.
Three: invoice and reimbursement. For company usage, OpenAI issues US-style invoices that Chinese finance departments universally reject. No VAT invoice (增值税发票) means no accounting entry, which means someone has to front the money personally and expense it under a different category. This is a continuous headache for any incorporated business.
These three problems compound. Most Chinese developers either use a foreign card plus VPN (fragile), buy top-ups via a third-party service (opaque pricing and rug-pull risk), or give up and stick to domestic models (quality gap is real).
Comparison of Common Solutions
| Solution | Cost | Risk | Invoice | Recommended |
|---|---|---|---|---|
| Foreign card + VPN | Medium | Medium (bans, card rejection) | None | Individuals, short term |
| Friend pays for you | Low | High | None | Not recommended |
| Taobao/Xianyu resellers | Medium-high | Very high (fraud, account theft) | None | Not recommended |
| USDT to official APIs | High (FX + fees) | Medium (OTC required) | None | Crypto-literate users |
| API gateway (e.g. Router One) | Low (near-list price) | Low | Yes | Most developers |
The first three work around the problem rather than solving it. USDT is viable for those already using crypto, but not every developer has an OTC account. An API gateway separates "identity and payment" from "API usage": you pay a Chinese entity in RMB, the gateway holds the foreign credit cards and accounts with upstream providers, and your traffic routes through its endpoint. You skip the foreign card requirement, skip the network access issues, and get a Chinese VAT invoice for your finance team.
Topping Up Router One with WeChat Pay or Alipay
Router One is an OpenAI-compatible LLM gateway that accepts WeChat Pay, Alipay, Stripe, and USDT. The individual-developer flow takes about 30 seconds:
Step 1: Register. Visit router.one and sign up with email. No credit card or phone verification required.
Step 2: Go to billing. In the dashboard, find the "Deposit" or "Billing" entry.
Step 3: Enter amount and select method. Minimum 10 CNY (roughly $1.50). Pick WeChat Pay or Alipay — a QR code appears.
Step 4: Scan to pay. Use your WeChat or Alipay app to scan. Balance arrives in seconds.
Step 5: Create API key. Go to "API Keys" and generate a key (format sk-...).
Step 6: Use it.
from openai import OpenAI
client = OpenAI(
base_url="https://api.router.one/v1",
api_key="sk-your-router-one-key",
)
response = client.chat.completions.create(
model="anthropic/claude-sonnet-4",
messages=[{"role": "user", "content": "Hello"}],
)
Claude Code, Codex CLI, and most OpenAI-compatible tools accept a custom base URL; specific configuration is covered in our Claude Code setup guide and Codex CLI alternative.
Team and Company Scenarios
The above flow assumes an individual developer. Teams and companies have extra considerations:
- Large one-time top-ups — Router One supports single deposits at 100,000+ CNY, though for risk-control reasons you should contact support first to confirm limits.
- Shared balance across members — one account can host multiple API keys, each named and budgeted separately. Billing is unified at the account level.
- VAT invoices — Router One issues 增值税普通发票 (and 专票 in certain cases, by support request). Submit company name and tax number through the dashboard; electronic invoice arrives in 1–3 business days. This is the concrete difference from official US-based API accounts.
- Reconciliation — the dashboard provides full transaction history export (top-ups, API charges, balance changes) as CSV for finance review.
- Budgets and alerts — per-project or per-key spending caps with automatic request rejection at 100% and email alerts at thresholds like 50%, 80%. Essential for teams, since one agent bug can burn hundreds of dollars overnight.
Alternatives: USDT and Stripe
Router One also supports USDT (TRC20 / ERC20) and Stripe:
- USDT — best for users already holding crypto. Anonymous and no FX quota, but requires an OTC account to acquire.
- Stripe — best for users with foreign cards or USD invoicing needs. Standard international Invoice, with ~2.9% + $0.30 fee.
- WeChat Pay / Alipay — best for most individual and company developers in China. Instant, zero learning curve, VAT invoice available.
FAQ
How long does top-up take? WeChat Pay and Alipay are instant (seconds). If 1 minute passes without credit, confirm the payment completed in your wallet app, then contact support with the transaction ID.
Minimum amount? 10 CNY (~$1.50). Start small to verify the flow before larger top-ups.
Can I get a refund? Consumed balance (used by API calls) is not refundable. Unused balance can be refunded with roughly 1–2% processing fee.
Are my prompts stored? No. Router One proxies requests in real time without persisting prompt or response content. Only metadata (model, tokens, latency, cost) is logged for billing and observability.
Supported models? OpenAI (GPT-4.1, o-series, Codex), Anthropic (Claude 4 Opus/Sonnet/Haiku), Google (Gemini 2.5 Pro/Flash), Mistral, DeepSeek, Meta Llama. Full catalog at router.one/models.
Configuring Claude Code and Codex
Once payment is handled, connecting your AI coding tools is just environment variables:
# Claude Code
export ANTHROPIC_BASE_URL=https://api.router.one
export ANTHROPIC_API_KEY=sk-your-router-one-key
export ANTHROPIC_AUTH_TOKEN=sk-your-router-one-key
# OpenAI Codex CLI
export OPENAI_BASE_URL=https://api.router.one/v1
export OPENAI_API_KEY=sk-your-router-one-key
Add them to ~/.zshrc or ~/.bashrc for persistent configuration. Detailed steps and measured latency from Chinese ISPs are in the Claude Code in China guide and the Codex China landing page.
Conclusion
Paying for OpenAI, Claude, and Gemini APIs is not unsolvable for Chinese developers — the official channels just did not bother to make it easy. An API gateway like Router One abstracts the identity-and-payment layer, letting you top up mainstream LLM APIs with WeChat Pay or Alipay and get proper Chinese VAT invoices. The remaining work is a one-line base URL change in your code. Model selection and developer experience are identical to official endpoints.
To see the full model catalog, go to models; for the end-to-end scenario pages, see the WeChat Pay for LLM API landing page and the Claude Code in China landing page.