LLM API Gateway User Guide (English)
English setup reference for API endpoints, model IDs, billing, and supported clients.
Welcome to Guishu LLM Token. This is the English customer edition of the Guishu Token guide. It has been rewritten for international users, with practical setup steps, clear pricing language, and a more familiar support style for UK and overseas customers.
1. Start Here
In most cases, you only need three things: your API key, the compatible API base URL, and the model ID you want to use.
| What you need | Use this |
|---|---|
| Buy or top up | Account center |
| Check balance and usage | https://token.gpt-agent.cc |
| International default API base URL | https://api.llm-token.cn/v1 |
| Chat completions endpoint | https://api.llm-token.cn/v1/chat/completions |
| Image generation endpoint | https://api.llm-token.cn/v1/images/generations |
| Image web app | https://image2.gpt-agent.cc/ |
Configuration tip: Some clients ask for a Base URL, while others ask for the full endpoint. If a client asks for a Base URL, use https://api.llm-token.cn/v1. If it asks for a complete chat endpoint, use https://api.llm-token.cn/v1/chat/completions.
Recommended First Test
curl https://api.llm-token.cn/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-sonnet-4-6",
"messages": [
{"role": "user", "content": "Please reply with one short sentence."}
]
}'2. Buying, Balance and Billing
Self-service
- Purchase and top up at Account center.
- Check remaining balance and usage at token.gpt-agent.cc.
- Use the same API key across supported models.
Enterprise support
- For larger top-ups, invoices, enterprise onboarding or dedicated routing, contact your account manager or support contact.
- Enterprise teams can request account-level guidance, deployment support and usage monitoring.
实时模型与计费口径
输入、输出和缓存输入可能采用不同单价。图片、视频或按次计费的模型请核对计费单位。模型 ID 和价格以实时定价页为准，实际扣费以账单记录为准。
Price Monitoring Skill
If your team uses OpenClaw or an automation bot, you can monitor the public model and pricing table for changes. The monitor detects new models, removed models, rate changes and price changes, then produces a notification that can be sent to a team chat or automation channel.
clawhub install feishu-public-table-monitorAlready installed? Update to the latest version:
clawhub --workdir ~/.openclaw/workspace update feishu-public-table-monitor --force3. Model Catalogue
从实时列表复制完整模型 ID。不同 API Key 或线路的访问范围可能不同，接入前请确认所选模型。
| 模型 | 厂商 | 价格 | 上下文 | 图片能力 | 复制操作 |
|---|---|---|---|---|---|
| 阿里云 / Qwen | 3 倍率 · ¥1.2 / 百万 Tokens | 1M | 支持图片 | ||
| 阿里云 / Qwen | 4 倍率 · ¥1.6 / 百万 Tokens | 1M | 支持图片 | ||
| 阿里云 / Qwen | 10 倍率 · ¥4 / 百万 Tokens | 1M | 文本输入 | ||
| 阿里云 / Qwen | 20 倍率 · ¥8 / 百万 Tokens | 1M | 支持图片 | ||
| 阿里云 / Qwen | 5 倍率 · ¥2 / 百万 Tokens | 1M | 支持图片 | ||
| 美团 / LongCat | 1 倍率 · ¥0.4 / 百万 Tokens | 1M | 文本输入 | ||
| MiniMax | 1 倍率 · ¥0.4 / 百万 Tokens | 1M | 支持图片 | ||
| 阶跃星辰 / StepFun | 1 倍率 · ¥0.4 / 百万 Tokens | 256K | 支持图片 | ||
| 字节跳动 / Doubao | 3 倍率 · ¥1.2 / 百万 Tokens | 128K | 支持图片 | ||
| 小米 / MiMo | 4 倍率 · ¥1.6 / 百万 Tokens | 1M | 文本输入 | ||
| 小米 / MiMo | 4 倍率 · ¥1.6 / 百万 Tokens | 1M | 支持图片 | ||
| 小米 / MiMo | 2 倍率 · ¥0.8 / 百万 Tokens来源有差异 | 1M | 支持图片 | ||
| DeepSeek | 7.5 倍率 · ¥3 / 百万 Tokens来源有差异 | 1M | 文本输入 | ||
| DeepSeek | 2.5 倍率 · ¥1 / 百万 Tokens来源有差异 | 1M | 文本输入 | ||
| 月之暗面 / Kimi | 25 倍率 · ¥10 / 百万 Tokens | 1M | 支持图片 | ||
| 智谱 / Z.ai | 8 倍率 · ¥3.2 / 百万 Tokens | 1M | 支持图片 | ||
| 智谱 / Z.ai | 8 倍率 · ¥3.2 / 百万 Tokens | 1M | 文本输入 | ||
| 智谱 / Z.ai | 8 倍率 · ¥3.2 / 百万 Tokens来源有差异 | 1M | 支持图片 | ||
| DeepSeek | 2.5 倍率 · ¥1 / 百万 Tokens来源有差异 | 1M | 支持图片 | ||
| 阿里巴巴 | 输入¥0.8/ 百万 Tokens输出¥4/ 百万 Tokens缓存¥0.08/ 百万 Tokens | — | 待确认 |
3 倍率 · ¥1.2 / 百万 Tokens
4 倍率 · ¥1.6 / 百万 Tokens
10 倍率 · ¥4 / 百万 Tokens
20 倍率 · ¥8 / 百万 Tokens
5 倍率 · ¥2 / 百万 Tokens
1 倍率 · ¥0.4 / 百万 Tokens
1 倍率 · ¥0.4 / 百万 Tokens
1 倍率 · ¥0.4 / 百万 Tokens
3 倍率 · ¥1.2 / 百万 Tokens
4 倍率 · ¥1.6 / 百万 Tokens
4 倍率 · ¥1.6 / 百万 Tokens
2 倍率 · ¥0.8 / 百万 Tokens
7.5 倍率 · ¥3 / 百万 Tokens
2.5 倍率 · ¥1 / 百万 Tokens
25 倍率 · ¥10 / 百万 Tokens
8 倍率 · ¥3.2 / 百万 Tokens
8 倍率 · ¥3.2 / 百万 Tokens
8 倍率 · ¥3.2 / 百万 Tokens
2.5 倍率 · ¥1 / 百万 Tokens
输入¥0.8/ 百万 Tokens输出¥4/ 百万 Tokens缓存¥0.08/ 百万 Tokens
数据口径更新于 2026-09-25T11:10:04.896Z
从实时列表复制完整模型 ID。不同 API Key 或线路的访问范围可能不同，接入前请确认所选模型。
4. API Access for Developers
Regional API Base URLs
All regions use the same globally routed Base URL. Your API key, model IDs and request format stay the same wherever you are.
| Customer location | Recommended API Base URL | Notes |
|---|---|---|
| All regions | https://api.llm-token.cn/v1 | Globally routed entry (nodes in East China, South China, Hong Kong, Singapore, the US, and Germany) that automatically picks the fastest route. |
If access is ever unstable, retry with the same Base URL — routing will pick the next fastest node automatically.
Chat Completions
curl https://api.llm-token.cn/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.5",
"messages": [
{"role": "user", "content": "Summarise this project in three bullet points."}
]
}'Image Generation
curl https://api.llm-token.cn/v1/images/generations \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-image-2",
"prompt": "A clean modern product banner for an AI platform, blue and white colour palette, professional SaaS style",
"response_format": "url",
"n": 1,
"aspect_ratio": "16:9"
}'4K image output: Check the current model list for available routes, capabilities and billing units.
Image tip: URLs returned by image APIs may be temporary. Download or store generated images if they are needed for a website, campaign or production workflow.
Common Request Fields
| Field | Purpose |
|---|---|
model | The model ID. Copy it exactly from the model list. |
messages | Used for chat-style requests. |
prompt | Used for image generation or simple text-style tasks. |
response_format | For images, url is usually easiest for testing and download. |
aspect_ratio | Image ratio, for example 1:1, 16:9 or 9:16. |
5. Tool Setup Guides
Universal setup pattern: choose an OpenAI-compatible or custom provider, enter the Guishu API base URL, paste your API key, then choose the model ID.
| Tool or guide | English setup instructions |
|---|---|
| Image generation | Use https://api.llm-token.cn/v1/images/generations. Choose gpt-image-2 according to the required image workflow. Start with one image per request for testing. |
| OpenClaw | Open settings, add an OpenAI-compatible provider, set Base URL to https://api.llm-token.cn/v1, paste your API key, then select a model such as claude-sonnet-4-6, MiniMax-M3 or gpt-5.5. For long-context work, confirm the client context window is not capped locally. |
| Tencent Cloud | When using a Tencent Cloud function, workflow or server-side app, store the Guishu API key as an environment variable and call the compatible endpoint from your backend. Avoid placing keys in frontend code. |
| Cherry Studio | Add a custom OpenAI-compatible provider. Use Base URL https://api.llm-token.cn/v1, paste your API key, and enter the model ID exactly as listed. |
| Claude Code | Use the supplied one-click script where available, or configure a compatible model route through your local provider settings. Recommended default: claude-sonnet-4-6. Use claude-opus-4-8 only for high-value work. |
| Codex | Configure an OpenAI-compatible endpoint with https://api.llm-token.cn/v1. Use coding-focused models such as gpt-5.6-terra, gpt-5.5, qwen3.7-plus or deepseek-v4-pro. |
| Hermes | Configure the gateway or client provider with the Guishu base URL and API key. Recommended models are claude-sonnet-4-6 for high concurrency and MiniMax-M3 for long-context workflows. |
| gpt-image-2 image model | Use the image generation endpoint and set model to gpt-image-2. This model is best for higher-quality images, design assets and polished marketing visuals. |
| WorkBuddy | Add a custom AI provider. Set the Base URL to https://api.llm-token.cn/v1, paste the API key, then choose a model. If a saved model name fails, copy a model ID directly from this guide. |
| OpenCode | Use an OpenAI-compatible provider. For fast coding choose gpt-5.6-terra or step-3.7-flash. For complex refactors choose deepseek-v4-pro, gpt-5.5 or claude-opus-4-8. |
| VS Code | For extensions that support custom OpenAI-compatible providers, enter the Base URL, API key and model ID. Test with a short prompt before starting a long coding session. |
| Chatbox | Select a custom OpenAI-compatible source. Use Base URL https://api.llm-token.cn/v1, paste your API key, and select a model. If the client asks for the full endpoint, use https://api.llm-token.cn/v1/chat/completions. |
| Trae | Use the custom provider or OpenAI-compatible configuration. Choose a coding model and run a small test prompt before applying it to an existing repository. |
| Cursor | Use custom model/provider settings where available. Add the Guishu base URL and API key, then select a coding model. Keep high-rate models for complex planning or review rather than every autocomplete request. |
| Image web app | Open image2.gpt-agent.cc, choose the image model, enter your prompt and download the result after generation. |
Recommended Tool Defaults
| Scenario | Recommended models |
|---|---|
| Daily automation and high concurrency | claude-sonnet-4-6, gpt-5.6-terra, MiniMax-M3 |
| Complex reasoning and high-quality writing | gpt-5.6-sol, qwen3.8-max, deepseek-v4-pro |
| Top-tier long-running Agent tasks | claude-opus-5 |
| Fast code iteration | gpt-5.6-terra, LongCat-2.0, qwen3.7-plus |
| Long-context documents | MiniMax-M3, qwen3.7-plus, LongCat-2.0, deepseek-v4-pro, kimi-k3, glm-5.2, glm-5.3, glm-5.3-flash |
| Image generation | gpt-image-2, gpt-image-2-4k |
6. Image Prompting Guide
For image generation, a clear prompt normally produces a better result. A useful structure is:
A [ratio] [use case] image of [subject], in [style], with [colour palette], [composition], [important constraints].Example:
A 16:9 marketing banner for an AI platform, clean SaaS style, blue and white colour palette, a subtle data visualisation background, premium and professional, no text.For social media or article covers, use 16:9. For app icons and avatars, use 1:1. For mobile posters, use 9:16.
7. Troubleshooting
| Issue | What to check |
|---|---|
| Authentication failed | Check that the header is exactly Authorization: Bearer YOUR_API_KEY. Make sure there are no extra spaces before or after the key. |
| Model not found | Copy the model ID exactly from the model list. Model names are case-sensitive and hyphen-sensitive. |
| Client rejects the URL | If the client asks for a Base URL, use https://api.llm-token.cn/v1. If it asks for a full endpoint, use https://api.llm-token.cn/v1/chat/completions. |
| Image URL expired | Download generated images promptly. Temporary image URLs are not intended as permanent asset hosting. |
| Context window appears too small | Some local clients cap context length in their own settings. Increase the local context window where supported, especially in OpenClaw or similar Agent tools. |
| Slow streaming or timeout | Image generation and long Agent tasks can take longer than normal chat. Increase client timeout settings and test with a shorter request first. |
| Balance seems wrong | Check usage at token.gpt-agent.cc. Billing is based on actual model rate and token usage recorded by the system. |
OpenClaw 16k Context Fix
If OpenClaw reports a rate limit or context issue while the selected model supports a much larger context window, the local client may still be capped at 16k. Check the OpenClaw model or gateway configuration and raise the context setting to match the model you intend to use.
Support Case Style
When reporting an issue to support, please include:
- Your API key name or account identifier, but never paste the full secret key in a public chat.
- The model ID you used.
- The client or tool name.
- The Base URL or endpoint you configured.
- The exact error message and approximate time of the request.
8. Coverage of the Chinese Source Documents
This English guide consolidates and localises the main Chinese guide and its linked child documents. The content has been rewritten for international customers rather than translated word-for-word.
| Chinese source area | Where it appears in this English guide |
|---|---|
| Main Guishu Token user guide | Sections 1, 2, 4 and 7 |
| Model list | Section 3 |
| Model pricing and billing | Section 2 and Section 3 |
| MiniMax image generation guide | Section 4 and Section 6 |
| OpenClaw setup guide | Section 5 and OpenClaw context fix in Section 7 |
| Tencent Cloud setup guide | Section 5 |
| Cherry Studio setup guide | Section 5 |
| Claude Code deployment guide | Section 5 |
| Codex deployment guide | Section 5 |
| Hermes deployment guide | Section 5 |
| gpt-image-2 image model guide | Section 4 and Section 6 |
| WorkBuddy setup guide | Section 5 |
| OpenCode setup guide | Section 5 |
| VS Code setup guide | Section 5 |
| Chatbox setup guide | Section 5 |
| Trae setup guide | Section 5 |
| Cursor setup guide | Section 5 |
| Web image model guide | Section 5 and Section 6 |
| FAQ and customer cases | Section 7 |
Need help? For enterprise onboarding, invoicing, larger top-ups, dedicated routing or team deployment support, please contact your Guishu account manager or customer support contact.