你的卡片上有 3 个值Your card has 3 values Endpoint · Key · Model

配置任何工具都只做一件事:把这 3 个值填进去。下面每一页都是照抄就能用的完整步骤。Setting up any tool means one thing: entering these 3 values. Every page below is a copy-and-paste walkthrough.

接入地址 EndpointEndpoint
https://api.thetoken.biz
你的 KeyYour Key
YOUR_KEY
你的模型名 ModelYour Model
YOUR_MODEL
模型名要原样照抄卡片。不要自己改成 gpt-4o 这类名字,会直接报 403 model_not_allowed
唯一的例外是 Claude 系工具(/anthropic 接口):那里填 claude-* 也可以,网关会自动指向你卡片上的模型。
Copy the model name exactly as printed. Do not substitute names like gpt-4o — they return 403 model_not_allowed.
The one exception is Claude-family tools (the /anthropic surface): a claude-* name works there too, and the gateway points it at the model on your card.

找到你要用的工具Find your tool

左边点对应的一项,照着做即可。只需要配一个,不用全配 —— 所有工具共用同一把 Key、同一份余额。Pick the matching item in the left sidebar and follow it. You only need one — every tool shares the same key and the same balance.

你用的工具Your tool 难度Effort 配置里填的接入地址(Base URL)Base URL to enter
Chatbox
聊天软件chat app
最简单easiest https://api.thetoken.biz/openai
软件会自动补 /v1/chat/completionsthe app appends /v1/chat/completions
Cherry Studio
聊天软件chat app
最简单easiest https://api.thetoken.biz/openai
软件会自动补 /v1/chat/completionsthe app appends /v1/chat/completions
Claude Cowork
需开开发者模式needs Developer Mode
中等medium https://api.thetoken.biz/anthropic
⚠ 结尾没有 /v1no /v1 at the end
Claude Code · Claude SDK 中等medium https://api.thetoken.biz/anthropic
⚠ 结尾没有 /v1,多写会 404no /v1 — adding it causes a 404
Codex CLI · opencode · OpenAI SDK 中等medium https://api.thetoken.biz/openai/v1
这里结尾要有 /v1the /v1 is required here
ChatGPT 官方 Appofficial app 用不了not usable 官方 App 不能改接入地址 → 请看左边「ChatGPT App」页的替代方案The official app cannot use a custom endpoint → see the "ChatGPT App" page for alternatives
Key 填在哪一个框里、叫什么名字,各工具不同 —— 每一页都写好了,照抄即可。Where the key goes and what it is called differs per tool — each page spells it out, ready to copy.

用之前先知道两件事Two things to know

回答是一次性出现的,不会逐字打出来。为了保证计费准确,网关要等模型答完并记账成功才把内容发给你。看起来「卡住不动」几秒是正常的。Answers appear all at once, not word by word. The gateway waits for the model to finish and for metering to commit before releasing the text. A few seconds of "nothing happening" is normal.
单次请求有上限:约 30 秒 / 约 2 MB,最多 8192 个输出 token。让它一次写太长的东西会失败。把任务拆小就好。One request is capped at roughly 30 s / 2 MB and 8192 output tokens. Asking for one very long answer will fail — split the task into smaller ones.

opencode CLI openai/v1

① 打开(或新建)配置文件① Open (or create) the config file
macOS / Linux
~/.config/opencode/opencode.json
Windows
C:\Users\<你的用户名your name>\.config\opencode\opencode.json
② 粘贴以下内容并保存② Paste this and save
opencode.json
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "thetoken": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "The Token",
      "options": {
        "baseURL": "https://api.thetoken.biz/openai/v1",
        "apiKey": "YOUR_KEY"
      },
      "models": {
        "YOUR_MODEL": { "name": "YOUR_MODEL", "tool_call": true }
      }
    }
  }
}
③ 替换占位符③ Replace the placeholders
  • YOUR_KEY 换成卡片上的 Key。Replace YOUR_KEY with the key on your card.
  • 全部 3 处 YOUR_MODEL 换成卡片上的模型名。Replace all 3 occurrences of YOUR_MODEL with the model name on your card.
④ 重启 opencode,然后验证④ Restart opencode, then verify
shell
opencode run --model thetoken/YOUR_MODEL "What is 2+2?"
⚠ 改完配置必须重启 opencode 才生效。⚠ You must restart opencode for the change to take effect.

Codex CLI openai/v1

① 打开(或新建)配置文件① Open (or create) the config file
macOS / Linux
~/.codex/config.toml
Windows
C:\Users\<你的用户名your name>\.codex\config.toml
② 粘贴以下内容并保存② Paste this and save
config.toml
model = "YOUR_MODEL"
model_provider = "thetoken"

[model_providers.thetoken]
name = "The Token"
base_url = "https://api.thetoken.biz/openai/v1"
wire_api = "chat"
http_headers = { "api-key" = "YOUR_KEY" }
③ 替换占位符③ Replace the placeholders
  • YOUR_MODEL 换成卡片上的模型名。Replace YOUR_MODEL with the model name on your card.
  • YOUR_KEY 换成卡片上的 Key。Replace YOUR_KEY with the key on your card.
④ 验证④ Verify
shell
codex exec --skip-git-repo-check "What is 2+2?"
wire_api 请保持 "chat"。改成 "responses" 也能跑,但多轮对话偶尔会断。Keep wire_api set to "chat". "responses" also works but can drop multi-turn sessions.

Claude Code anthropic

全部配置只改一个文件不要登录 Anthropic 账号,不需要官方订阅,不需要安装任何代理软件。The entire setup is one file. Do not log in to an Anthropic account, no subscription is needed, and no proxy software is required.

① 找到(或新建)这个文件① Find (or create) this file
macOS / Linux
~/.claude/settings.json
Windows
C:\Users\<你的用户名your name>\.claude\settings.json
文件或 .claude 文件夹不存在,自己新建即可。用记事本 / VS Code 打开就行。If the file or the .claude folder does not exist, just create it. Any text editor works.
② 粘贴以下内容并保存② Paste this and save
settings.json
{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.thetoken.biz/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_KEY",
    "ANTHROPIC_MODEL": "YOUR_MODEL",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "YOUR_MODEL",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "YOUR_MODEL",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "YOUR_MODEL",
    "ANTHROPIC_SMALL_FAST_MODEL": "YOUR_MODEL",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}
文件里已经有内容时:不要整个覆盖,只把 "env": { … } 这一整块加到最外层大括号里,其余原样保留(注意上一项末尾要补一个英文逗号)。If the file already has content: do not overwrite it. Add the whole "env": { … } block inside the outermost braces and keep the rest (remember the comma after the previous entry).
③ 替换两个占位符③ Replace the two placeholders
  • YOUR_KEY 换成卡片上的 Key。Replace YOUR_KEY with the key on your card.
  • 全部 5 处 YOUR_MODEL 都换成卡片上的同一个模型名(用编辑器的「全部替换」最快)。Replace all 5 occurrences of YOUR_MODEL with the same model name from your card (use "Replace all").
  • 保存。检查引号和逗号都是英文半角。Save. Make sure all quotes and commas are plain ASCII.
④ 关掉终端窗口,重新打开,验证④ Close the terminal, reopen it, and verify
shell
claude -p "What is 2+2?"
能返回 4 就配好了。之后想换模型或换 Key,只改这个文件、重开终端即可。If it answers 4, you are done. To change model or key later, edit this file and reopen the terminal.

最容易踩的 3 个坑The 3 usual mistakes

地址结尾多写了 /v1必须是 …/anthropic。Claude Code 会自己补 /v1/messages,多写就变成 /v1/v1/messages404Adding /v1 to the URL. It must be …/anthropic. Claude Code appends /v1/messages itself; with the extra segment you get /v1/v1/messages404.
写成了 ANTHROPIC_API_KEY必须用 ANTHROPIC_AUTH_TOKEN,否则会触发 Claude Code 自己的账号校验流程。Using ANTHROPIC_API_KEY. Use ANTHROPIC_AUTH_TOKEN instead, otherwise Claude Code runs its own account validation flow.
只填了 ANTHROPIC_MODELClaude Code 还会用「小模型」跑起标题、命令建议等后台任务。5 个模型变量必须全部填成同一个值,否则聊天正常但状态栏一直报 403Only setting ANTHROPIC_MODEL. Claude Code also uses a "small" model for titles and suggestions. All 5 model variables must be the same value, or chat works while the status line keeps showing 403.

其他说明Also worth knowing

这个文件里明文存着你的 Key。个人电脑上没问题,但不要提交到 git —— 尤其是项目目录下的 .claude/settings.jsonThis file stores your key in plain text. Fine on a personal machine, but keep it out of git — especially a project-level .claude/settings.json.
Claude Agent SDK读同一批变量,上面配好就能直接用。Reads the same variables — nothing extra to do.
Claude Desktop不支持自定义接入地址(只能登录 Anthropic 账号),无法使用本服务。想要图形界面请看「桌面聊天软件」页。Does not support a custom endpoint (account login only), so it cannot be used here. For a GUI, see the Desktop apps page.
thinking(扩展思考)不生效,会被忽略;其余功能包括工具调用都正常。thinking (extended thinking) is ignored; everything else, including tool use, works normally.

Claude Cowork anthropic

Cowork 是 Claude 桌面应用里的办公助手模式。它自带一个「开发者模式」,可以把它接到我们这里 —— 全程点鼠标填表格,不用写任何配置文件Cowork is the assistant mode inside the Claude desktop app. Its Developer Mode lets you point it here — all clicking and form-filling, no config files.

只有 WindowsmacOS 版支持,Linux 不支持。请先不要登录 Anthropic 账号 —— 停在登录界面,先做下面第 ① 步。Supported on Windows and macOS only. Do not sign in to an Anthropic account yet — stay on the login screen and do step ① first.
① 打开开发者模式① Turn on Developer Mode
Windows
左上角应用菜单 → Help → Troubleshooting → Enable Developer ModeTop-left app menu → Help → Troubleshooting → Enable Developer Mode
macOS
顶部菜单栏 Help → Troubleshooting → Enable Developer ModeMenu bar → Help → Troubleshooting → Enable Developer Mode
完全退出应用再重新打开,菜单栏才会多出一个 Developer 菜单。Fully quit and reopen the app — only then does a Developer menu appear.
② 菜单栏 Developer → Configure third-party inference② Menu bar: Developer → Configure third-party inference
弹出的窗口左侧有 ConnectionIdentity & Models 等几栏。先点 ConnectionThe dialog has a sidebar with Connection, Identity & Models and more. Start with Connection.
③ 在 Connection 里填这 4 项③ Fill in these 4 fields under Connection
Connection 类型type
Gateway (Anthropic-compatible)(Anthropic-compatible)
Gateway base URL
https://api.thetoken.biz/anthropic
Gateway API key
YOUR_KEY
Gateway auth scheme
bearer
Gateway extra headers 留空,不用填。auth schemex-api-key 也能用,但建议先用 bearerLeave Gateway extra headers empty. x-api-key also works as the auth scheme, but start with bearer.
④ Identity & Models:留空④ Identity & Models: leave it empty
模型列表不用填。应用会自己来问我们有哪些模型,然后在模型下拉框里显示 claude-sonnet-4-5
如果它坚持要你填,就填 claude-sonnet-4-5
You do not need to list models. The app asks us and shows claude-sonnet-4-5 in the picker.
If it insists on a value, enter claude-sonnet-4-5.
⑤ 点 Apply locally,应用会自动重启⑤ Click Apply locally — the app restarts itself
重启后的登录界面上会多出一个 Cowork on 3P(第三方推理)的入口 —— 点它进去。不要点登录 Anthropic 账号。After the restart the sign-in screen offers Cowork on 3P (third-party inference) — click that. Do not sign in to an Anthropic account.
⑥ 验证⑥ Verify
新建一个会话,发一句「2+2 等于几」。能回答就配好了。Start a session and ask "What is 2+2?". If it answers, you are done.

你会注意到的几件事Things you will notice

模型名显示成 claude-sonnet-4-5 是正常的。Cowork 只认 Anthropic 自家的模型名,所以我们对它统一报这个名字。实际跑的、计费的都是你卡片上的模型Seeing claude-sonnet-4-5 in the picker is expected. Cowork only accepts Anthropic's own model names, so that is the name we advertise. What actually runs — and what gets billed — is the model on your card.
地址结尾不要写 /v1必须是 …/anthropic,应用会自己补后面的路径。多写会全部 404。No /v1 at the end. It must be …/anthropic; the app appends the rest. The extra segment makes everything 404.
回答是整段一次性出现的,不会逐字打字。等几秒是正常的。Answers appear all at once, not typed out. A few seconds of waiting is normal.
要检查配置有没有被读进去:Help → Troubleshooting → Copy Managed Configuration Report(Key 会自动打码)。To check what the app actually loaded: Help → Troubleshooting → Copy Managed Configuration Report (secrets are redacted).

ChatGPT App N/A

OpenAI 官方的 ChatGPT 应用(Windows / macOS / 手机 / 网页版)无法使用本服务。它没有「填入接入地址和 Key」的地方 —— 只能用 OpenAI 自己的账号登录。这不是我们的限制,任何第三方接口都接不进去。OpenAI's official ChatGPT app (Windows / macOS / mobile / web) cannot be used with this service. It has no field for a custom endpoint and key — it only signs in to an OpenAI account. This is not our limitation; no third-party endpoint can be used there.
别被「开发者模式」误导。ChatGPT 里的 Developer mode(设置 → 应用 / 连接器 → 高级)是用来接 MCP 连接器的,跟换接入地址、换模型完全无关,打开了也没用。Don't be misled by "Developer mode". ChatGPT's Developer mode (Settings → Apps / Connectors → Advanced) is for adding MCP connectors. It has nothing to do with changing the endpoint or the model.
想要类似 ChatGPT 的体验,用这些替代Use one of these instead
替代软件Alternative 适合谁Who it suits
Chatbox界面最像 ChatGPT,安装完填 3 个框就能聊。推荐一般办公用户从这个开始Closest to ChatGPT's feel; install, fill 3 fields, start chatting. Best starting point for most people.
Cherry Studio功能更多(知识库、翻译、多模型切换),愿意多点几下就选它。More features (knowledge base, translation, multiple providers) if you don't mind a few more clicks.
Codex CLIOpenAI 官方的命令行工具,支持自定义接入地址。适合写代码的人。OpenAI's own command-line tool — it does support a custom endpoint. For people who write code.
左边侧栏点对应的一项,照着做即可。Pick the matching page in the left sidebar and follow it.

Chatbox openai

最省事的一个:装好软件,填 3 个框,就能像用 ChatGPT 一样聊天。The easiest option: install, fill in 3 fields, chat like you would in ChatGPT.

① 下载安装① Download and install
官网 chatboxai.app → 下载对应版本(Windows / macOS / 手机都有,免费)→ 安装 → 打开。Go to chatboxai.app → download your version (Windows / macOS / mobile, free) → install → open.
② 设置 → 模型提供方,选「OpenAI API」② Settings → Model provider → choose "OpenAI API"
左下角齿轮图标是设置。找不到「OpenAI API」就选「添加自定义提供方」,类型选 OpenAIThe gear icon at the bottom left opens Settings. If "OpenAI API" is not listed, choose "Add custom provider" and set the type to OpenAI.
③ 填这 3 个框③ Fill in these 3 fields
API 域名 / API HostAPI Host
https://api.thetoken.biz/openai
API 密钥 / API KeyAPI Key
YOUR_KEY
模型 / ModelModel
YOUR_MODEL
API 域名结尾不要写 /v1Chatbox 会自己在后面接上 /v1/chat/completions。如果软件里有单独的「API 路径」框,保持默认的 /v1/chat/completions 不要改。Do not add /v1 to the API host. Chatbox appends /v1/chat/completions itself. If there is a separate "API Path" field, leave its default /v1/chat/completions alone.
④ 添加模型④ Add the model
  • 点「获取模型列表 / 刷新」,列表里应该出现你卡片上的模型 —— 选中它。Click "fetch / refresh model list" — the model on your card should appear. Select it.
  • 列表拉不出来也没关系:直接手动输入模型名,原样照抄卡片,一个字符都不要改。If the list does not load, just type the model name manually, copied exactly from your card.
⑤ 保存,新建对话,发一句「2+2 等于几」⑤ Save, start a new chat, ask "What is 2+2?"
能回答就配好了。回答会整段一次性出现,不会逐字打字,这是正常的。If it answers, you are done. The reply appears all at once rather than word by word — that is expected.

配不通时先看这里If it does not connect

报 404。八成是地址写重了。软件最终请求的地址必须正好是 https://api.thetoken.biz/openai/v1/chat/completions —— 出现 /v1/v1 就是「API 域名」里多写了 /v1404. Almost always a doubled path. The final request URL must be exactly https://api.thetoken.biz/openai/v1/chat/completions — a /v1/v1 means you added /v1 to the host.
报 401。Key 没填上,或者复制时带了空格 / 换行。删掉重贴一次。401. The key is missing, or was pasted with a stray space or newline. Clear the field and paste again.
报 403 model_not_allowed。模型名不对,必须原样照抄卡片,不能自己改成 gpt-4o 之类。403 model_not_allowed. Wrong model name — copy it exactly from your card; names like gpt-4o will not work.

Cherry Studio openai

功能比 Chatbox 多(知识库、翻译、画图入口等),配置也只是填两个框 + 手动加一个模型。More features than Chatbox (knowledge base, translation and more); setup is still two fields plus adding one model.

① 下载安装① Download and install
官网 cherry-ai.com → 下载对应版本(Windows / macOS / Linux,免费)→ 安装 → 打开。Go to cherry-ai.com → download your version (Windows / macOS / Linux, free) → install → open.
② 左下角设置 → 模型服务 → 添加自定义服务商② Bottom-left Settings → Model providers → add a custom provider
名字随便起(比如 The Token),提供商类型必须选 OpenAIName it anything (e.g. The Token). The provider type must be OpenAI.
③ 填这 2 个框③ Fill in these 2 fields
API 地址API host
https://api.thetoken.biz/openai
API 密钥API key
YOUR_KEY
API 地址结尾不要写 /v1,也不要以斜杠结尾。Cherry Studio 会自己接上 /v1/chat/completionsNo /v1 and no trailing slash. Cherry Studio appends /v1/chat/completions itself.
④ 添加模型④ Add the model
  • 点页面左下角的「管理」按钮,会自动列出你的 Key 能用的模型。Click Manage at the bottom left — it lists the models your key can reach.
  • 点模型右边的 + 把它加进模型列表。不点 + 的话,聊天时的模型下拉框里是找不到它的。Click the + next to the model to add it to the list. Without this it will not appear in the chat model picker.
  • 列表拉不出来就手动添加:模型 ID 原样照抄卡片。If the list does not load, add it by hand: the model ID is copied exactly from your card.
⑤ 打开右上角的启用开关,再点「检测」⑤ Flip the enable switch (top right), then click Check
这一步最容易漏。右上角的开关不打开,这个服务商就是「未启用」状态,聊天时的模型列表里根本看不到它。This is the most commonly missed step. If the top-right switch is off, the provider stays disabled and its models never show up in chat.
⑥ 新建对话,选中模型,发一句「2+2 等于几」⑥ New chat → pick the model → ask "What is 2+2?"

配不通时先看这里If it does not connect

报 404 或看到 /v1/v1「API 地址」里多写了 /v1。删掉即可。
如果删掉还是 404,改填完整地址并在结尾加一个 #https://api.thetoken.biz/openai/v1/chat/completions# —— # 是 Cherry Studio 的约定,表示「就用这个地址,别再自动拼接」。
404, or you see /v1/v1. You added /v1 to the host — remove it.
If it still 404s, enter the full path and end it with #: https://api.thetoken.biz/openai/v1/chat/completions# — in Cherry Studio the # means "use this URL as-is, do not append anything".
模型列表里找不到模型。要么忘了点模型右边的 +,要么忘了打开右上角的启用开关。The model is missing from the chat picker. Either you did not click the + next to it, or the provider's enable switch is still off.
报 401。Key 填错,或复制时带了空格。注意 Cherry Studio 支持逗号分隔多个 Key —— 别不小心粘进了逗号。401. Wrong key, or a stray space. Note Cherry Studio treats commas as multiple keys — make sure none slipped in.

OpenAI SDK openai/v1

已有的 OpenAI 代码一行都不用改逻辑,只把 base_urlapi_key 换掉即可。Existing OpenAI code needs no logic changes — only swap base_url and api_key.

① 安装① Install
shell
pip install openai        # Python
npm install openai        # Node.js
② Python② Python
python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.thetoken.biz/openai/v1",
    api_key="YOUR_KEY",
)
r = client.chat.completions.create(
    model="YOUR_MODEL",
    messages=[{"role": "user", "content": "hi"}],
)
print(r.choices[0].message.content)
curl (REST)(REST)
shell
curl https://api.thetoken.biz/openai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"YOUR_MODEL","messages":[{"role":"user","content":"hi"}]}'

Anthropic SDK / REST anthropic

SDK 的 base_url 结尾不要写 /v1(SDK 自己会补 /v1/messages);curl 直连时才写完整路径。另外 max_tokens 是必填项,不填会 400Do not put /v1 in the SDK base_url (the SDK appends /v1/messages); only curl uses the full path. Also, max_tokens is required — omitting it returns 400.
Python (anthropic SDK)(anthropic SDK)
python
from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.thetoken.biz/anthropic",
    api_key="YOUR_KEY",
)
r = client.messages.create(
    model="YOUR_MODEL",
    max_tokens=1024,
    messages=[{"role": "user", "content": "hi"}],
)
print(r.content[0].text)
curl (REST)(REST)
shell
curl https://api.thetoken.biz/anthropic/v1/messages \
  -H "x-api-key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"YOUR_MODEL","max_tokens":1024,"messages":[{"role":"user","content":"hi"}]}'
TypeScript 的 @anthropic-ai/sdk 一样:new Anthropic({ baseURL: "https://api.thetoken.biz/anthropic", apiKey: "YOUR_KEY" })TypeScript @anthropic-ai/sdk is the same: new Anthropic({ baseURL: "https://api.thetoken.biz/anthropic", apiKey: "YOUR_KEY" }).

出错了怎么办Troubleshooting

你看到的提示What you see怎么办What to do
404 或「找不到」/ "not found"地址写错了。Claude Code 用 …/anthropic不带 /v1),其他工具用 …/openai/v1要带 /v1)Wrong URL. Claude Code uses …/anthropic (no /v1); everything else uses …/openai/v1 (with /v1)
401 missing_api_keyKey 没填上,或填错了位置。回到对应页面重新核对The key is missing or in the wrong field — recheck against your tool's page
403 model_not_allowed模型名不对,必须原样照抄卡片。(仅 Claude 系工具可以填 claude-*Wrong model name — copy it exactly from your card. (Only Claude-family tools may use a claude-* name.)
400 model_alias_ambiguous你的 Key 可用多个模型,claude-* 无从判断该用哪个 → 改填卡片上的真实模型名Your key can reach several models, so a claude-* name cannot pick one → use the real model name from your card
Claude Code 能聊天,但一直闪 403Claude Code chats fine but keeps flashing 4035 个模型变量没填全,回到 Claude Code 页第 ③ 步Not all 5 model variables are set — see step ③ on the Claude Code page
Claude Code / cowork 报“不是 Anthropic 模型”Claude Code / cowork says "not an Anthropic model"这类工具只认 Anthropic 自己的模型名 → 模型格填 claude-sonnet-4-5,网关会自动指向你卡片上的模型These tools only accept Anthropic's own model names → enter claude-sonnet-4-5 and the gateway points it at the model on your card
403 quota_exhausted
403 insufficient_credit
额度用完了 → 联系管理员充值Out of quota → ask the admin to top up
403 key_expiredKey 到期 → 联系管理员续期Key expired → ask the admin to renew
400 (自己写代码时)(when writing your own code)Anthropic 接口漏了 max_tokens,它是必填的The Anthropic endpoint is missing max_tokens — it is required
长任务做到一半断了A long task stops halfway超出单次 30 秒 / 2 MB 上限。把任务拆小再试It exceeded the 30 s / 2 MB per-request limit — split the task up
回答不是一个字一个字出来的The answer is not typed out word by word正常现象。为了准确计费,内容会等模型答完后一次性返回Expected. For accurate metering the text is released only after the model finishes
改了配置没生效Config change had no effect关掉终端 / 客户端重新打开。配置只在启动时读一次Close and reopen the terminal / app — config is read only at startup
还是不行?把你看到的完整报错文字发给管理员,不要发你的 Key。Still stuck? Send the full error text to your admin — never send your key.