Resource points now convert to tokens: one plan covers cloud resources and AI calls

Building an app with AI — the annoying part isn't writing code. Writing code you can get excited about; reconciling the bill, not so much.

How much are the cloud resources? How much does one AI model call cost? Each item bought separately, managed separately — before the project even launches, the bill already makes your head spin.

What this upgrade solves​

CloudBase has upgraded its billing: resource points now directly offset AI model token usage.

Resource points become the unified billing unit, with cloud resource consumption and AI calls charged from the same pool. You only manage one quota.

Resource points as universal currency: one pool pays for cloud and AI tokens​

What are resource points​

In one line: resource points = CloudBase's unified currency. Database, cloud functions, storage, traffic and AI calls — all cloud resources — can be converted interchangeably. Not enough? Top up a resource-point pack or switch to pay-as-you-go.

Charge order: plan → resource pack → pay-as-you-go.

1,000 points = ¥1, 10,000 points = ¥10.

New users: pick a plan and start​

New users pick any full-stack Plan in the console, resource-point billing is on by default, and you can start developing.

We offer three plans — pick one by scale, no self-assembly needed:

• Personal ¥19.9/month (limited-time offer) 40,000 resource points, roughly 2 million database calls or ~190 million tokens* (DeepSeek-V4-Flash input), plus 4GB storage. Enough for practice and a first trial.
• Standard ¥199/month, 330,000 resource points, roughly 16 million database calls or ~1.5 billion tokens*, plus 300GB storage traffic, with WeChat Pay. Fits independent developers and one-person companies.
• Enterprise ¥999/month, 1,500,000 resource points, roughly 75 million database calls or ~6.9 billion tokens*, plus a dedicated database + WeCom integration. Fits teams and enterprise apps.

*Token conversion note: estimated using DeepSeek-V4-Flash as the direct model and 80% input cache hit rate (cache hit 20 points / million tokens, miss 1,000 points / million tokens). Actual cache hit rates vary by scenario — for reference only; see the official resource-point pricing docs for the exact numbers.

After development, you get a publicly accessible online URL you can share with anyone, not just a local preview.

Existing users: balance and validity fully preserved​

We've prepared a smooth migration for existing users; existing Token Resource Packs remain valid:

• Old-plan users: after switching to resource-point billing in the console, Token Resource Pack balance converts to resource points automatically, with validity preserved — not a single point lost.
• Resource-point-plan users: existing Token Resource Packs keep working; the unconsumed portion before August 31, 2026 rolls over to the general resource-point pack automatically, no manual action needed.

Switch path: console → Plan usage page → switch to resource-point billing, done in one step.

How tokens pay with resource points​

When calling an AI model, input and output convert to tokens and are charged directly from resource points — no more buying token packs.

Each model's unit price is listed as "resource points / million tokens." Supported mainstream models include GLM-5.1, Kimi-K2.6, MiniMax-M3, Qwen3.5-Plus and more; see the AI console → text-generation model for the full list.

Running within the plan is a discount on the LLM too​

Resource points themselves are discounted within the plan, so using in-plan resource points for AI is effectively a discount on the model.

Let's do the math:

DeepSeek-V4-Flash's official list price is ¥1/million tokens input, ¥2/million tokens output. If you're on the Personal plan (¥19.9/month, 40,000 points), each resource point is worth ¥0.00498 — half the pay-as-you-go price.

At that discounted rate, DeepSeek-V4-Flash effectively costs ¥0.5/million tokens input, ¥1/million tokens output.

A conversation with 5k input + 2k output tokens: list price ¥0.009, but in-plan it's only ¥0.0045. That's two calls for a penny.

You don't decide the model in advance​

Enable the models you need in the AI console, then specify them via the model parameter when calling — the charge path is identical. Switching models doesn't affect balance management.

In closing: how to use tokens​

Scenario one: configure a custom model in WorkBuddy​

1. Open a task → model config → add custom model
2. Get the environment ID: console → view environment → copy environment ID
3. Get the Base URL and API Key: console home "AI dev tools → use tokens in AI tools → copy Base URL and API Key", or "console → AI → text-generation model → copy Base URL and API Key"
4. Click save config
5. Done — start a conversation in the workspace and verify the model works

Whichever model you pick, copy it accurately.

Cursor, Cline and Claude Code configure the same way — Provider set to OpenAI Compatible, fill in Base URL and API Key.

Scenario two: call directly from a mini program​

Streaming calls, fit for real-time conversation and long-text generation:

const model = wx.cloud.extend.AI.createModel("cloudbase");

const res = await model.streamText({
  data: {
    model: "deepseek-v4-flash",
    messages: [{ role: "user", content: "介绍一下李白" }],
  }
});

for await (const text of res.textStream) {
  console.log("文本片段:", text);
}

Switch models by changing only the model field:

model: "minimax-m2.7"  // MiniMax
model: "glm-5v-turbo"  // GLM
model: "kimi-k2.6"     // Kimi

Billing should serve development. The goal of this upgrade is one thing: let you stop worrying about quotas and spend time on what matters — shipping the product.