· reAPI Team
Token Plan or Token 中转站? Buying AI Model Access in China
Token plans from Alibaba Cloud and Volcengine vs token 中转站 API relays vs pay-per-use gateways: prices, limits and what to check before you buy model access.
Developers in China usually buy AI model access in one of three ways: a cloud vendor's subscription, such as Alibaba Cloud's Token Plan; a "token 中转站", a reseller that relays requests to models behind its own base URL; or a pay-per-use API gateway with published per-model prices. In our view, the token plan gives the clearest terms, the 中转站 often looks cheapest, and the gateway sits between them.
This guide uses Alibaba Cloud's own Token Plan documentation, checked on November 4, 2026, to show what such a subscription actually includes and what it costs per Credit, then sets out what to check before trusting any relay.
What a token plan is
Alibaba Cloud describes Token Plan as an AI model subscription measured in Credits, covering text and multimodal models plus their built-in tools, and compatible with tools such as Claude Code, Cursor, Qwen Code and OpenClaw. The team edition provides a base URL compatible with both OpenAI and Anthropic interfaces.[1][2] Volcengine sells comparable subscriptions under the names Coding Plan and Agent Plan in its Ark platform.[3]
Alibaba's personal edition has four tiers, all at promotional prices:[1]
| Tier | Promotional price | List price | Credits per month | Concurrent agents |
|---|---|---|---|---|
| Lite | ¥39 | ¥60 | 11,500 | 1 to 2 |
| Essential | ¥79 | ¥120 | 25,500 | 2 to 3 |
| Standard | ¥139 | ¥180 | 45,000 | 3 to 4 |
| Pro | ¥499 | ¥600 | 180,000 | 6 to 8 |
Extra usage packs cost ¥100 a month for 20,000 Credits, up to five at a time, and require an active plan. The team edition is sold per seat: ¥150 (list ¥198) for 25,000 Credits, ¥550 (list ¥698) for 100,000, and ¥1,398 for 250,000, with shared packs at ¥5,000 for 625,000 Credits.[1][2]
Dividing Credits by price shows which tier is better value. At the promotional prices, Lite gives about 295 Credits per yuan, Essential about 323, Standard about 324 and Pro about 361. A usage pack gives 200 Credits per yuan, so if you regularly buy packs, moving up a tier is cheaper than topping up. Credits also pay for built-in tools such as web search, web extraction and a code interpreter when you call models through the Responses API.[1]
Three details matter more than the headline price. When a plan's monthly Credits and packs run out, service pauses until the next subscription month. Both editions currently work only in Alibaba's North China 2 (Beijing) region. And there are night discounts: from 22:00 to 08:00, Qwen3.8 Max and Flash calls use Credits at 40% of the normal rate and DeepSeek V4 and V4.1 models at 50%.[1] If your workload can run overnight, that changes the value of a tier considerably.
What a token 中转站 is
A token 中转站 sells access to well-known models through its own endpoint. You change your base URL and key, and the station forwards requests and bills you in its own units. That convenience is real, especially for models you cannot easily buy directly.
The risks are also real, and none of them is visible in the price list. Before paying, check:
- Which model actually answers. A model name in a request is not proof of what runs behind it. Test with prompts whose answers you can verify, and compare output quality and token counts with the vendor's own API.
- The billing unit. "Credits", "points" or a multiplier on the official price need converting to dollars or yuan per million tokens before they can be compared.
- What happens on failure. Find out whether failed or timed-out requests are charged, and whether a balance can be refunded.
- Where your data goes. Prompts pass through the station's servers. Read its privacy terms, and keep secrets and customer data out until you have.
- Whether it is authorized. Check whether the model vendor's terms allow the station to resell access, since an account it relies on can be suspended without warning.
A relay that cannot answer these questions plainly is cheap for a reason you will find out later.
Pay-per-use gateways
A gateway publishes a price per model and bills per request from one balance, with no monthly tier to use up. reAPI is one: its model pages list per-token or per-second prices, its credits are fixed at $0.001 each, and its pricing panel states that you pay only for completed generations.[4] For comparison with Token Plan's models, reAPI lists DeepSeek V4.1 Flash at $0.40 per million input tokens and $1.30 per million output tokens.[4]
A gateway is not automatically cheaper than a subscription. A plan's flat monthly price wins if you use most of its Credits, especially at night; a gateway wins when usage is uneven or you need models outside the plan's list.
Choosing between them
- Steady daily coding in China, inside the plan's model list: a subscription, sized to the Credits you actually use.
- Bursty or experimental usage: a pay-per-use gateway, so idle months cost nothing.
- A model the official clouds do not sell you: a relay or gateway, after running the checks above.
- Team use with budgets: Alibaba's team edition offers seat assignment, per-member usage analysis and caps on shared packs.[2]
FAQ
What is the cheapest Alibaba Cloud Token Plan?
The Lite tier at a promotional ¥39 per month, with 11,500 Credits and one to two concurrent agents; its list price is ¥60.[1]
Can I use a token plan with Claude Code or Cursor?
Alibaba lists Claude Code, Cursor, Qwen Code, Qoder and OpenClaw as compatible tools.[1]
Is a token 中转站 legal?
It depends on the model vendor's terms and how the station obtains access. Read the vendor's terms for resale before relying on one for production work.
Token plan or token 中转站 in practice
A token plan is the transparent choice when its model list and region fit your work. A token 中转站 can fill gaps, but only after you have verified the model, the billing unit and the data handling. For per-request billing with published prices across chat, image and video models, compare reAPI's model catalog with the plan you are considering, and pick the token plan, relay or gateway whose real cost per finished task is lowest.
References
- Alibaba Cloud Model Studio (阿里云百炼). Token Plan personal edition overview. Retrieved November 4, 2026 from help.aliyun.com/zh/model-studio/token-plan-personal-overview
- Alibaba Cloud Model Studio (阿里云百炼). Token Plan team edition overview. Retrieved November 4, 2026 from help.aliyun.com/zh/model-studio/token-plan-team-overview
- Volcengine. Ark documentation navigation (Agent Plan, Coding Plan). Retrieved November 4, 2026 from docs.volcengine.com/docs/ark
- reAPI. DeepSeek V4.1 Flash model page and pricing panel. Retrieved November 4, 2026 from reapi.ai/models/deepseek-v4-1-flash