DeepSeek, Qwen, Kimi and MiniMax behind one OpenAI-compatible endpoint — served from the Shantou cable landing station, invoiced in USD, no Chinese payment rails and no mainland business entity required. DeepSeek、通义千问、Kimi、MiniMax，统一走一个 OpenAI 兼容接口。 从汕头国际海缆登陆站出海，美元开票，不需要支付宝微信，也不需要你在中国注册公司。
Chinese labs ship some of the best price-performance in the industry. The friction is everything around the model: paying a mainland entity, getting an invoice your finance team accepts, and reaching an endpoint that is fast from where you actually are. 中国几家大模型的性价比是全行业顶尖的。卡人的从来不是模型， 是模型外面那一圈事：怎么付钱给中国主体、怎么拿到你们财务认的发票、 以及从你所在的地方访问那个接口到底快不快。
USD invoicing against a bank transfer or card. No Alipay, no WeChat Pay, no mainland corporate account, no one on your team explaining a CNY wire to finance. 美元发票，电汇或刷卡。不用支付宝、不用微信支付、 不用在中国开对公户，也不用跟财务解释一笔人民币汇款是怎么回事。
Shantou's cable landing station carries 52% of China's international bandwidth, with 32.7 ms to Singapore. If your users are in ASEAN, this is the short path. 汕头国际海缆登陆站承载全国 52% 的国际出口带宽，到新加坡 32.7 毫秒。 如果你的用户在东盟，这是最短的一条路。
OpenAI-compatible. Swap base_url and model name; your SDK, your retries, your streaming code all keep working. Move between models without a rewrite. OpenAI 兼容。改 base_url 和模型名就行，你的 SDK、重试逻辑、 流式代码全都不用动。想换模型不用重写。
A single contracting entity, a normal service agreement, and an invoice with a line item. Not a prepaid balance on a console you can't read. 一个签约主体、一份正常的服务协议、一张有明细的发票。 不是某个你读不懂的控制台里的一笔预充值余额。
Per-key metering, per-model breakdown, exportable. You should be able to reconcile your bill against your own logs without asking us. 按 key 计量、按模型拆分、可导出。你应该能拿自己的日志核我们的账单， 不用问我们。
Streaming ASR at $0.25 per audio hour and neural TTS well under the Western incumbents. If you run voice at scale, start here. 流式语音识别 0.25 美元/音频小时，语音合成远低于欧美同类。 如果你的量在语音上，从这里开始。
Published in USD. These are indicative list prices for the early-access cohort — committed volume gets better. We would rather show you real numbers now and hold to them than hide behind "contact sales for pricing". 以美元标价。这是给早期接入客户的参考价，有保底量的价格更好。 我们宁可现在就把真实数字摆出来并守住它，也不想写「价格请联系销售」。
If your code already talks to OpenAI, it already talks to us. 你的代码只要能跟 OpenAI 说话，就能跟我们说话。
# Python — openai SDK from openai import OpenAI client = OpenAI( base_url="https://tokenpier.sinogenomics.com/v1", # ← 1 api_key="tp-your-key-here", # ← 2 ) r = client.chat.completions.create( model="deepseek-v3.2", messages=[{"role": "user", "content": "Hello"}], stream=True, )
Models, rough monthly volume, and where your users sit. Takes one form below. 用哪些模型、每月大概多少量、用户在哪。填下面那个表就行。
A test key with trial credit, plus written pricing for your volume. 一把带试用额度的测试 key，外加针对你的量的书面报价。
Run your own evals and latency tests from your own region before committing anything. 在你自己的区域跑你自己的评测和时延测试，测完再决定要不要签。
Postpaid against metered usage, or prepaid if your finance team prefers it. 按计量后付费；你们财务喜欢预付也可以。
This matters more than the price, and most resellers are vague about it. We won't be. 这件事比价格重要，而且大部分转卖的人说不清楚。我们说清楚。
Your request crosses into a physically isolated processing zone at Shantou, the model runs there, and only the result comes back. This is the structure the pilot policy is built on — data in, inference in-zone, result out. If your workload cannot leave your own jurisdiction, this product is not for you, and we will tell you so rather than take the order. 你的请求进入汕头一个物理隔离的处理区，模型在那里跑，只有结果回传。 试点政策的结构就是这样——数据进来、区内推理、结果出去。 如果你的业务数据依法不能离开你所在的司法辖区，这个产品不适合你， 我们会直接告诉你，而不是先把单子接下来。
Metadata needed to bill you: timestamp, key, model, token counts, latency, status. Prompt and completion bodies are not retained by us for training or resale. Upstream model providers apply their own retention policies, and we will name the provider behind every model you use — you should not have to guess. 只记开账单需要的元数据：时间、key、模型、Token 数、时延、状态。 提示词和回复正文我方不留存、不拿去训练、不转卖。 上游模型厂商有各自的留存策略，你用的每个模型背后是哪家我们都会明确告诉你， 不该让你去猜。
Often you should — if an international card works for you and latency from your region is fine, OpenRouter is a good product and we will say so. We are a better fit when you need USD invoicing from a single counterparty, when your finance or legal team needs a real service agreement, when you are running heavy speech workloads, or when your users are in ASEAN and the extra round-trip matters. 很多时候你确实应该用——如果国际信用卡能付、时延也够，OpenRouter 是个好产品， 这话我们照说。我们更合适的场景是：你需要一个签约主体开美元发票、 你们财务法务需要一份正式服务协议、你的量压在语音上、 或者你的用户在东盟而那一跳往返时延对你有意义。
We are a distributor sourcing capacity through a Chinese supply partner, not a first-party lab. We will tell any prospect exactly that, in writing, before they sign anything. Anyone in this market who implies otherwise is worth a second look. 我们是通过中国的供应合作方拿产能的分销方，不是模型厂商本身。 任何客户签约前，我们都会白纸黑字把这句话讲清楚。 这个市场里暗示自己是厂商直营的，你值得多看两眼。
Honest answer for an early-access product: today we have a single supply route, so an upstream outage is your outage. We are not going to pretend otherwise with a made-up uptime number. Do not put a critical single-provider dependency on us yet — run us alongside your existing provider, measure, and expand only when you are satisfied. 早期产品，实话实说：目前只有一条供应链路，上游中断就是你中断。 我们不会编一个可用性数字来糊弄。现在别把关键路径单点押在我们身上—— 跟你现有的供应商并行跑，测够了再加量。
Yes. Early-access accounts get trial credit and no commitment. We would rather you benchmark us properly and walk away than sign and be disappointed. 可以。早期接入的账号有试用额度，不要求承诺量。 我们宁可你认真压测完之后觉得不行走人，也不想你签了之后失望。
Chinese model prices have been falling, not rising — there were six price cuts across the major Chinese labs in the first half of 2026. Our contracts pass reductions through. If an upstream price ever rises, we give notice before it reaches your invoice. 中国的模型价格一直在降不是在涨——2026 上半年主要几家一共降了六次。 我们的合同把降价传导给你。万一上游真涨价，会在账单之前先通知你。
A real person reads every one of these. If we are not the right fit for your workload we will say so — that answer is cheaper for both of us than a bad contract. 每一条都有人看。如果我们不适合你的场景，我们会直说—— 这个答案对双方来说都比一份糟糕的合同便宜。