mofcloud

How to use Chinese AI APIs from outside China: DeepSeek, Qwen, Doubao, Kimi, GLM and MiniMax (2026)

Where overseas teams sign up for DeepSeek, Qwen, Doubao, Kimi, GLM, MiniMax and Huawei MaaS: international platforms, regions, currency and data residency.

MofCloud

Nov 19, 2026 · 8 min read

To use Chinese AI models from outside China, sign up on each company’s international platform, not its China site. As of November 2026, each of the major Chinese model makers has one: DeepSeek’s API platform, Alibaba Cloud Model Studio for Qwen, BytePlus ModelArk for ByteDance’s Seed (Doubao) models, Kimi’s API platform, Z.ai for GLM, MiniMax’s international platform, and Huawei Cloud’s international MaaS. These platforms price in USD and accept international accounts. What differs between them is where your requests are processed, which matters more than price if you handle customer data.

This guide covers the entry point for each provider, what changes between China and international sites, where inference runs, and the shortcuts (aggregators and your existing cloud) if you’d rather not open seven new accounts.

Where do you sign up for each Chinese AI API?

Here’s the international entry point for each, with where it documents inference as running. Checked November 19, 2026, against each provider’s own docs.

Model family International platform China-site equivalent Currency Where inference runs (as documented)
DeepSeek DeepSeek API platform (api.deepseek.com) Same platform USD, prepaid balance Personal data processed and stored in the PRC (privacy policy)
Qwen (Alibaba) Alibaba Cloud Model Studio Bailian on aliyun.com USD You pick: Singapore, Hong Kong, Tokyo, Virginia, Frankfurt, or Beijing
Seed / Doubao (ByteDance) BytePlus ModelArk Volcengine Ark USD Johor, Malaysia (ap-southeast-1) or Dublin (eu-west-1)
Kimi (Moonshot) Kimi API platform (platform.kimi.ai) platform.moonshot.cn USD, before tax Not stated on the pricing page
GLM (Zhipu) Z.ai open.bigmodel.cn USD Not stated on the pricing page
MiniMax MiniMax platform (platform.minimax.io) platform.minimaxi.com USD Not stated on the pricing page
DeepSeek, GLM, Qwen on Huawei Huawei Cloud MaaS (international) Huawei Cloud China MaaS Pay-per-use CN-Hong Kong region (model release notes)

In one line: each model family has an international door; the doors differ mainly in where the compute sits.

Most of these platforms expose an OpenAI-compatible Chat Completions endpoint, so switching is usually a base URL, an API key and a model name. BytePlus ModelArk, for example, documents https://ark.ap-southeast.bytepluses.com/api/v3 for its Asia-Pacific region.

Qwen outside China: Alibaba Cloud Model Studio

Qwen’s international route is Model Studio on the Alibaba Cloud international site. You choose a region, and that choice drives three things:

  • Data location. Alibaba says request data is stored in the region you pick, and requests are sent to inference nodes “within the service deployment scope.” Hong Kong, Tokyo, Virginia and Frankfurt each offer a “Global” scope and a local one (Hong Kong, Japan, US, EU). If you need data to stay in the EU or the US, pick the local scope, not Global.
  • API keys. Keys are region-specific and can’t be used across regions. A Beijing key won’t work against Singapore.
  • Price. Each region has its own price list, and Beijing is often cheaper than Singapore. Singapore comes with a free quota for new users, valid for 90 days.

One deprecation to plan for: Alibaba says the older DashScope domain (dashscope-intl.aliyuncs.com for Singapore) won’t get new features after September 30, 2026, and recommends the workspace-specific {WorkspaceId}.{region}.maas.aliyuncs.com domain.

If you’re new to Alibaba Cloud, we cover the China site vs international site split, billing and cost tools in Alibaba Cloud cost management for international teams.

Doubao and Seed outside China: BytePlus ModelArk

ByteDance sells its models internationally through BytePlus ModelArk, the counterpart of Volcengine Ark in China. As of November 19, 2026:

  • Regions: Johor (ap-southeast-1) for the full model list, and Dublin (eu-west-1) for a subset. The EU region list on the region availability page names seed-2-0-lite and offers online inference but not batch inference.
  • Spillover: BytePlus says requests to an endpoint go primarily to that endpoint’s region, but “some requests may be routed to inference resources in other regions.” Its data processing page names data centers in Malaysia, Indonesia and/or the EU/EEA.
  • Country list: BytePlus publishes the countries where its model service is available. The list, last updated April 21, 2026, doesn’t include the United States or mainland China. Check it before you build on ModelArk from the US.
  • Models: ByteDance’s Seed models (branded “Dola Seed” for the newest), plus hosted DeepSeek and GLM models. Some Volcengine models, such as Doubao Seed 2.1 Pro and Kimi K3.1 Pro, aren’t on the BytePlus list.

We compare the two storefronts model by model in BytePlus vs Volcengine.

DeepSeek outside China

DeepSeek runs one API platform for everyone, priced in USD on its English docs, with usage deducted from a prepaid balance (DeepSeek pricing). The catch is data location: DeepSeek’s privacy policy (last updated February 10, 2026) says it collects, processes and stores personal data in the People’s Republic of China. Data your own end users send through an app you build is excluded from that policy, because you’re the controller for it, but the API requests still go to DeepSeek.

If that’s a problem, use DeepSeek’s open-weight models through someone else’s infrastructure. BytePlus ModelArk lists DeepSeek V4 Pro at the same $1.32 / $3.96 per million tokens as DeepSeek’s own platform (BytePlus model pricing), and Huawei Cloud’s international MaaS serves DeepSeek models from Hong Kong. Western clouds and aggregators host DeepSeek models too (see below).

Takeaway: DeepSeek’s model is open-weight; DeepSeek’s API isn’t the only way to run it.

Kimi, GLM and MiniMax outside China

These three run separate international platforms with their own domains and USD price lists:

  • Kimi moved its international platform to platform.kimi.ai (the old platform.moonshot.ai redirects there). Pricing is pay-as-you-go, with tax calculated at checkout based on your jurisdiction. K2 models are closed to new users from October 31, 2026.
  • GLM is sold internationally as Z.ai, in USD. The China platform, open.bigmodel.cn, is in CNY, and the two price lists aren’t identical.
  • MiniMax uses platform.minimax.io internationally and platform.minimaxi.com in China.

None of the three states an inference location on its pricing page. If residency matters, read their privacy terms and data processing agreements before sending production data.

Huawei Cloud MaaS international

Huawei Cloud MaaS (ModelArts Studio) offers pay-per-use model inference on Huawei’s international site. Its release notes list DeepSeek V4 Pro, DeepSeek V4.1 Flash, GLM-5.3 and several Qwen models, all in the CN-Hong Kong region. In April 2026 Huawei said MaaS would become available in nine more countries, including Singapore, Thailand, Indonesia, Brazil, Mexico, Saudi Arabia, the UAE, South Africa and Türkiye (announcement). That’s availability for customers in those countries; the model list still shows Hong Kong as the serving region.

People search for Huawei MaaS pricing a lot, and we couldn’t find a public per-token price list on Huawei’s international site. We track the models and prices on our Huawei Cloud MaaS page.

What changes between a China site and an international site?

China site International site
Account Chinese real-name verification (ID or business license) International sign-up
Currency and tax CNY, prices often include VAT USD, usually before tax
Price Often lower; Qwen in Beijing, GLM on Bigmodel Same or higher for most models; DeepSeek is the same
Model list Often the full lineup, first Often a subset, sometimes later
Data location Mainland China Depends on provider and region
Docs and console Chinese first English

The temptation is to sign up on a China site because it’s cheaper. For a team outside China that usually doesn’t work: real-name verification needs Chinese ID or a Chinese business license, and data would sit in mainland China, which creates cross-border transfer questions we cover in Is your cloud bill safe with a FinOps tool?.

Shortcuts: aggregators and your existing cloud

You don’t have to open an account with every Chinese lab:

  • Model aggregators such as OpenRouter, SiliconFlow, Together AI and DeepInfra serve many Chinese open-weight models behind one key and one bill. You trade some price and model freshness for convenience, and data handling follows the aggregator’s terms, not the lab’s.
  • Your hyperscaler. AWS, Google Cloud and Azure host some open-weight Chinese models in their model catalogs. Spend lands on the bill you already have and inside your existing data agreements.
  • A Chinese cloud you already use. If you run on Alibaba Cloud international or Huawei Cloud, their model services also host DeepSeek and GLM.

Our take: for production, go direct (or through a cloud you already trust) for the one or two models you use most, and use an aggregator for experiments. Pick by where the data goes first and by price second.

How to keep the costs in one place

Every route above produces a different bill: a prepaid DeepSeek balance, a line on your Alibaba Cloud invoice, a BytePlus charge, an aggregator’s credit card statement. We cover how to pull these together in LLM API cost tracking and multi-cloud billing.

Mofis, our desktop app, reads model spend straight from Alibaba Cloud, Volcengine, Tencent Cloud, Huawei Cloud and Baidu bills, plus OpenAI, and shows it next to your AWS, Azure and Google Cloud costs in one currency. Before you buy, compare current per-token prices across all these providers on our LLM price comparison and Chinese LLM API pricing guide.

FAQ

Can I use DeepSeek API outside China?

Yes. DeepSeek’s API platform accepts international users and prices in USD. Its privacy policy says personal data is processed and stored in the PRC. If you need inference elsewhere, BytePlus ModelArk (Malaysia or Ireland) and Huawei Cloud MaaS (Hong Kong) host DeepSeek models.

How do I get a Qwen API key outside China?

Create an Alibaba Cloud international account, open Model Studio, choose a region (Singapore, or a Global or local scope in Hong Kong, Tokyo, Virginia or Frankfurt), and create an API key there. Keys only work in the region where you created them.

Can I use Doubao outside China?

Yes, through BytePlus ModelArk, ByteDance’s international platform. It offers ByteDance’s Seed models in USD from Malaysia, with some models in Ireland. Its published country list doesn’t include the United States, and some Volcengine models aren’t available on it.

Do I need a Chinese phone number or ID?

The international platforms are built for sign-ups from outside China, and price in USD. China sites such as Volcengine, Bigmodel and Alibaba Cloud’s China site require Chinese real-name verification.

Where is my data processed when I use Chinese AI APIs?

It depends on the platform and region. Alibaba Model Studio and BytePlus ModelArk let you choose regions outside mainland China. DeepSeek’s own platform processes data in the PRC. Kimi, Z.ai and MiniMax don’t state an inference location on their pricing pages, so check their data processing terms.


Using Qwen, Doubao or DeepSeek alongside AWS or Google Cloud? Mofis puts all of it in one view. It’s free for one person, with a 14-day Pro trial and no card required.