If you live outside mainland China, the hardest part of using DeepSeek is not the model — it is the checkout page. The official platform expects a Chinese phone number, and top-ups run through Alipay or WeChat Pay. Here is the workaround that does not involve borrowing someone's account.
base_url to https://api.novatoken.dev/v1. Same per-token list price, no Chinese identity required.
| Route | What it takes | Where it breaks |
|---|---|---|
| Official platform | Chinese phone number, real-name verification, Alipay / WeChat Pay top-up | Most international cards are simply not an option; verification is impractical abroad |
| Borrowed account or virtual CN card | A friend's account, or a resold wallet/virtual card | Account risk, recharge fees, and the same problem again next month |
| Card-billed gateway (NovaToken) | Email + any credit or debit card | Nothing recurring — one key, one balance, works for every model we serve |
NovaToken is the third option: we run DeepSeek-class models on our own infrastructure and partner inference capacity, expose them through the standard OpenAI API shape, and bill in USD. Per-token pricing matches the official list price — we are not adding a per-request fee on top.
base_url at https://api.novatoken.dev/v1 and use your new key.| Model | Best for | USD / 1M tokens |
|---|---|---|
| DeepSeek-V4-Flash | Reasoning, 1M context, agents | $0.30 in / $1.20 out |
| Qwen2.5-7B-Instruct | Fast chat, classification, tool calling | $0.10 in / $0.30 out |
| GLM-5.3-Flash | Bilingual work, summarisation | $0.15 in / $0.50 out |
| Kimi-K2.7-Code | Code, tool-heavy agents (262K context) | $1.20 in / $5.00 out |
Cache hits on DeepSeek-V4-Flash are billed at $0.006 / 1M input instead of $0.30 — a 50× discount applied automatically when the upstream reports a cache hit. Long system prompts, agents and coding assistants hit this constantly.
Real-world math: 1M input + 300K output on DeepSeek-V4-Flash is about $0.66. A small side project that sends a few thousand requests a month typically lands between $2 and $5. There is no subscription and nothing expires.
Yes — the same open-weight DeepSeek-V4 family, served on our infrastructure and partner capacity. Model names, context window, reasoning behaviour and streaming format are identical.
Yes. Change base_url and the key. Chat completions, streaming (SSE), tool/function calling and JSON mode all work.
No — the whole point is that you don't need them. Payment is by card in USD, which is what makes the service usable for people outside mainland China.
No. We store timestamps and token counts for billing only. Request bodies are not retained.
That is exactly the starting point. $5 of credits is roughly 25M input tokens on DeepSeek-V4-Flash — enough to build and test something real before committing more.
Yes. Every model is behind the same key and endpoint; you change the model string. If you don't want to choose at all, send "model": "auto" and each request is routed to a suitable model at that model's own rate.
Card payment, USD, no Chinese phone number, credits never expire.
Buy $5 of credits