三步开始

Three steps

注册、充值、建令牌，然后就可以调用了。

Register, add credit, create a key. Then start calling.

一把 key 只属于一个分组，分组决定它能用的模型。

A key belongs to exactly one group, and the group decides which models it can use.

接入地址

Endpoints

同一把 key、同一个域名，三种协议都可以用，都支持流式输出。

Same key, same host, three protocols. Streaming works on all of them.

Base URL 到底填哪个

Which base URL to enter

• OpenAI 系工具和 SDK（base_url / OPENAI_BASE_URL）：填 https://ainzy.net/v1
• Claude Code 和 Anthropic SDK（ANTHROPIC_BASE_URL）：填 https://ainzy.net，不带 /v1
• 鉴权统一用请求头 Authorization: Bearer sk-你的key；Anthropic 协议也接受 x-api-key

• OpenAI-style tools and SDKs (base_url / OPENAI_BASE_URL): https://ainzy.net/v1
• Claude Code and Anthropic SDKs (ANTHROPIC_BASE_URL): https://ainzy.net, without /v1
• Authenticate with the header Authorization: Bearer sk-YOUR_KEY; the Anthropic protocol also accepts x-api-key

不确定模型名时，先请求一次 /v1/models，列表里有的就能用。模型名区分大小写。

Not sure about a model name? Call /v1/models first. Anything in the list works. Model names are case-sensitive.

分组与模型

Groups & models

创建令牌时选择分组。价格以价格页为准，页面会按你的分组实时显示每百万 token 的单价。

Pick a group when creating a key. Prices live on the pricing page, which shows the per-million-token rate for your group.

工具接入

Tools

常用编程工具的配置方法。图形化切换请用 CC Switch，它会替你管理这些配置文件，并且能一键拉取你这把 key 能用的模型。

Configs for common coding tools. For a graphical switcher use CC Switch. It manages these config files for you and can fetch the models your key can use.

用 CC Switch 或任何工具导入国产模型的 key 之后，模型名也必须一起改掉。Claude Code 默认请求 claude- 开头的模型，国产模型分组里没有这个名字，不改就会报 model_not_found。在 CC Switch 里点一次「拉取模型」，或手动填 glm-5.3、kimi-k3。

After importing a Chinese-models key into CC Switch or any other tool, change the model name as well. Claude Code requests claude- models by default; those names do not exist in the Chinese-models groups, so you will get model_not_found. Click "fetch models" once in CC Switch, or type glm-5.3 or kimi-k3 manually.

Claude Code

令牌选「国模高缓存组」，把下面这段加进 ~/.claude/settings.json（Windows 在 %USERPROFILE%\.claude\settings.json）：

Create a key in the chinese-model-3 group and add this to ~/.claude/settings.json (on Windows: %USERPROFILE%\.claude\settings.json):

复制{
  "env": {
    "ANTHROPIC_BASE_URL": "https://ainzy.net",
    "ANTHROPIC_AUTH_TOKEN": "sk-你的key",
    "ANTHROPIC_MODEL": "glm-5.3",
    "ANTHROPIC_SMALL_FAST_MODEL": "glm-5.3-flash"
  }
}

Copy{
  "env": {
    "ANTHROPIC_BASE_URL": "https://ainzy.net",
    "ANTHROPIC_AUTH_TOKEN": "sk-YOUR_KEY",
    "ANTHROPIC_MODEL": "glm-5.3",
    "ANTHROPIC_SMALL_FAST_MODEL": "glm-5.3-flash"
  }
}

也可以直接设环境变量后运行 claude。模型名可换成同分组里的 kimi-k3、deepseek-v4-pro 等。

Setting the same environment variables before running claude also works. Swap the model for any other in the group, such as kimi-k3 or deepseek-v4-pro.

Codex CLI

令牌选国产模型分组，编辑 ~/.codex/config.toml：

Create a key in the Chinese-models group and edit ~/.codex/config.toml:

复制model = "glm-5.3"
model_provider = "ainzy"

[model_providers.ainzy]
name = "Ainzy"
base_url = "https://ainzy.net/v1"
env_key = "AINZY_API_KEY"
wire_api = "responses"

Copymodel = "glm-5.3"
model_provider = "ainzy"

[model_providers.ainzy]
name = "Ainzy"
base_url = "https://ainzy.net/v1"
env_key = "AINZY_API_KEY"
wire_api = "responses"

然后设环境变量 AINZY_API_KEY=sk-你的key 再运行 codex。

Then set AINZY_API_KEY=sk-YOUR_KEY in your environment and run codex.

Cursor / Cline / Roo Code / 其他 OpenAI 兼容工具

Cursor / Cline / Roo Code / other OpenAI-compatible tools

• 提供商选 OpenAI 或 OpenAI Compatible
• Base URL 填 https://ainzy.net/v1，API Key 填你的 sk- key
• 模型名手动填写分组里的模型，例如 deepseek-v4-flash、glm-5.3

• Choose OpenAI or OpenAI Compatible as the provider
• Base URL https://ainzy.net/v1, API key your sk- key
• Type the model name from your group, e.g. deepseek-v4-flash, glm-5.3

Cursor 覆盖 OpenAI Base URL 后请点「Verify」，通过即表示 key 和地址都对。

In Cursor, click "Verify" after overriding the OpenAI base URL. A pass means both the key and the URL are correct.

Cherry Studio

桌面聊天客户端，不写代码、只想直接对话时用它最省事。最快的办法是一键导入：到令牌页，点你那条令牌的「聊天」，选 Cherry Studio，会唤起软件并自动填好地址和 key（需要本机已安装 Cherry Studio）。

A desktop chat client — the easiest option when you just want to talk to a model instead of writing code. Fastest path is the one-click import: on the keys page, open the "Chat" menu on your key and pick Cherry Studio. It launches the app with the base URL and key already filled in (Cherry Studio must already be installed).

手动配置同样可以：

Manual setup works just as well:

• 设置 → 模型服务 → 添加提供商，类型选 OpenAI
• API 地址填 https://ainzy.net，API 密钥填你的 sk- key
• 在这个提供商下点「添加模型」，手动输入模型 ID，例如 glm-5.3、kimi-k3、deepseek-v4-pro
• 回到聊天页，在顶部模型下拉里选中刚添加的模型

• Settings → Model Providers → add a provider, type OpenAI
• API host https://ainzy.net, API key your sk- key
• Under that provider click "Add model" and type the model ID, e.g. glm-5.3, kimi-k3, deepseek-v4-pro
• Back on the chat page, pick the new model from the dropdown at the top

模型下拉里空着，说明模型还没添加——Cherry Studio 不会自动列出本站模型，必须手填一次。若提示 404，把 API 地址改成 https://ainzy.net/v1/（末尾带斜杠）再试。

An empty model dropdown just means no model has been added yet — Cherry Studio does not list ours automatically, so add one by hand. If you get a 404, change the API host to https://ainzy.net/v1/ (with the trailing slash) and retry.

代码示例

Code samples

把 sk-你的key 换成你的令牌、模型名换成分组里的模型，直接运行。

Replace sk-YOUR_KEY with your key and the model with one from your group, then run.

复制curl https://ainzy.net/v1/chat/completions \
  -H "Authorization: Bearer sk-你的key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "你好，介绍一下你自己"}],
    "stream": true
  }'

Copycurl https://ainzy.net/v1/chat/completions \
  -H "Authorization: Bearer sk-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Hi, introduce yourself"}],
    "stream": true
  }'

复制from openai import OpenAI

client = OpenAI(base_url="https://ainzy.net/v1", api_key="sk-你的key")

stream = client.chat.completions.create(
    model="glm-5.3",
    messages=[{"role": "user", "content": "你好，介绍一下你自己"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

Copyfrom openai import OpenAI

client = OpenAI(base_url="https://ainzy.net/v1", api_key="sk-YOUR_KEY")

stream = client.chat.completions.create(
    model="glm-5.3",
    messages=[{"role": "user", "content": "Hi, introduce yourself"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

复制import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://ainzy.net/v1", apiKey: "sk-你的key" });

const stream = await client.chat.completions.create({
  model: "kimi-k3",
  messages: [{ role: "user", content: "你好，介绍一下你自己" }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

Copyimport OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://ainzy.net/v1", apiKey: "sk-YOUR_KEY" });

const stream = await client.chat.completions.create({
  model: "kimi-k3",
  messages: [{ role: "user", content: "Hi, introduce yourself" }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

复制import anthropic

client = anthropic.Anthropic(base_url="https://ainzy.net", api_key="sk-你的key")

with client.messages.stream(
    model="glm-5.3",
    max_tokens=1024,
    messages=[{"role": "user", "content": "你好，介绍一下你自己"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Copyimport anthropic

client = anthropic.Anthropic(base_url="https://ainzy.net", api_key="sk-YOUR_KEY")

with client.messages.stream(
    model="glm-5.3",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hi, introduce yourself"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

复制curl https://ainzy.net/v1/responses \
  -H "Authorization: Bearer sk-你的key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "input": "用一句话解释什么是 API 网关"
  }'

Copycurl https://ainzy.net/v1/responses \
  -H "Authorization: Bearer sk-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "input": "Explain what an API gateway is in one sentence"
  }'

所有示例都用国产模型分组的 key 即可，Responses 接口同样支持国产模型。

All samples work with a Chinese-model-group key; the Responses endpoint supports the Chinese models too.

计费说明

Billing

• 按量计费，无月费。只按实际消耗的 token 扣费，输入、输出、缓存命中分别计价。
• 美元计价。余额单位是美元，价格页显示每百万 token 的美元单价。支付宝按 ¥7 = $1 换算。
• 分组决定折扣。同一模型在不同分组价格不同，令牌建好后分组即固定。
• 缓存命中更便宜。国产模型分组对命中缓存的输入 token 按缓存价计费，重复上下文越多越省。
• 失败请求不扣费。返回错误的请求不会计入账单。
• 不限单账号并发。没有每账号并发上限，高并发直接跑即可。
• 邀请返利 10%。在钱包页拿邀请链接，被邀请用户每笔充值的 10% 会在几分钟内进入你的邀请奖励，可划转到余额。

• Pay as you go, no monthly fee. Only tokens actually consumed are charged. Input, output and cache-hit tokens are priced separately.
• Priced in US dollars. Balance is in USD and the pricing page shows USD per million tokens. Alipay converts at ¥7 = $1.
• The group sets the discount. The same model costs differently in different groups, and a key's group is fixed once created.
• Cache hits cost less. In the Chinese-model groups, cached input tokens are billed at the cache rate. The more repeated context, the more you save.
• Failed requests are free. Requests that return an error are not billed.
• No per-account concurrency cap. Run as many parallel requests as you need.
• 10% referral rebate. Get your invite link on the wallet page. 10% of every top-up by an invited user lands in your referral balance within minutes and can be moved to your main balance.

每一次调用的模型、token 数、扣费都记录在使用日志，账单逐条可查。

Every call's model, token counts and charge are recorded in the usage log, so the bill is auditable line by line.

错误对照

Errors

常见报错和处理方法。

Common errors and what to do about them.

常见问题

FAQ