FAQ

Why does it say "insufficient quota" when my account balance is enough?

This is because token quota and account balance are separate:

• Token quota is only used to set the maximum usage limit
• Users can freely set their token quota
• Please check if your token quota is sufficient

What rate limits apply to model calls?

To maintain service stability, different rate limits apply to each model type:

TPM (Tokens Per Minute) is the number of tokens that can be processed per minute.

The TPM available to free models may vary with service load. If you encounter TPM throttling or related errors while using a paid model, please contact us and we will investigate and follow up.

Requests made after reaching the request limit may return 429 Too Many Requests or a similar error. Please wait for the rate-limit window to reset before trying again.

Please use the service responsibly. If abnormal usage is detected (such as unusually high-frequency requests, account sharing, etc.), we may restrict the account for 7 days to maintain service stability.

If you have any questions during use, please visit the About page to join our WeChat community group, and we will assist you promptly.

Why can't I use models marked as "free" before recharging, or why does CherryStudio show connection test errors?

Our services are provided through legitimate channels. While ensuring quality, we offer a series of models for users to quickly experience. Therefore, free models are only available to paid users. Users cannot access them before making a payment.

Why can't DeepSeek V4 Flash / Pro use Web Search in CherryStudio Agent?

Web Search in CherryStudio Agent is a server-side tool provided by cloud services. Its availability depends on whether the corresponding cloud service can support this capability reliably.

To ensure overall model service stability, CherryIN uses different cloud services to provide DeepSeek V4 Flash / Pro. Because support for Web Search varies across cloud services, CherryIN currently cannot provide this tool reliably for these models.

This is usually not caused by an API Key or CherryStudio configuration issue. If you need web search capabilities, we recommend configuring Exa MCP in CherryStudio.

Why does a model sometimes report an inaccurate version number?

If you ask a model, "What version are you?" or "What model are you?" and its answer does not match your API configuration, this is usually normal and does not indicate an integration problem.

A model generates answers from its training data, conversation context, and system prompt. By default, it cannot directly read the model name or version used for the current request. In addition, a model's official name and version number may only be finalized during or after training. When asked about its version, the model can only infer an answer from existing knowledge and may report an older, related, or inaccurate name.

This can happen with major model families, including Gemini, GPT, Claude, and Qwen.

Why can models in official apps usually report their versions correctly?

Official apps and websites often provide the current model name through a system prompt or other runtime information.

The model therefore appears to "know" its version, but the application has actually supplied that identity. Third-party clients and standard API requests usually do not inject this information, so the model's self-description may differ from the configured model.

How can I confirm which model I am using?

Use the model selected in the CherryIN console as the source of truth rather than the model's self-introduction. An incorrect self-reported version does not affect the model that is actually called or its output.

If you want to evaluate the model's capabilities further, use a third-party benchmark such as ToolCall-15. Standardized capability tests are more objective and informative than asking a model for its version number.

Can I get an invoice for API recharges in mainland China?

Unfortunately, this is not available at the moment. Our payment channels currently do not support this feature.

Where do I create an API Key, and which Base URL should I use?

Create an API Key on the Token Management page in the CherryIN console. For most OpenAI-compatible clients, use:

https://open.cherryin.net/v1

When using the built-in CherryIN provider in CherryStudio, you usually only need to enter the API Key. If a tool-specific guide provides different settings, follow that guide. See the OpenAI-Compatible Integration Guide for details.

Why do I get 401 Unauthorized?

A 401 response usually means authentication failed. Check the following:

• The API Key was copied completely
• There are no spaces or line breaks before or after the Key
• The request sends the Key as Authorization: Bearer <API_KEY>
• The client is not using a Key from another platform

Never share a complete API Key in chats, screenshots, or public code repositories.

Why do I get 404 or a model-not-found error?

First verify the Base URL and model name:

• Most OpenAI-compatible clients use https://open.cherryin.net/v1
• Use the complete model name in provider/model-name format, such as anthropic/claude-sonnet-4.5
• Sign in to the CherryIN console and confirm the currently available model ID on the Model List page

Some clients append /v1 automatically, so make sure the final request URL does not contain a duplicated path.

Can I use CherryIN with the OpenAI SDK or OpenAI-compatible tools?

Yes. Replace the SDK or tool's API Key with your CherryIN Key and set the Base URL to https://open.cherryin.net/v1. Model availability and parameters may differ, so use the complete model ID shown in the console. See the API Code Tutorial for examples.

How should I store my API Key securely?

• Use environment variables or a .env file included in .gitignore for local development
• Use cloud secrets or a dedicated secret manager in production
• Never put the Key in frontend code, public repositories, logs, or screenshots
• If you suspect a Key has leaked, disable it immediately and create a new one

How to fix the issue where CherryStudio conversations with Gemini Banana Pro model have a high probability of not generating images?

Here are some suggestions:

• In the CherryIN provider model endpoint, change google/gemini-3-pro-image-preview to the Gemini endpoint

• Conversation page settings: For image generation scenarios, disable streaming, and limit conversation context to 1-2