Models have limitations: they cannot access the latest information, are prone to hallucination, and struggle with precise calculations. Integrate plugins to address these limitations.
Introduction to plugins
A plugin is a collection of tools (APIs), each implementing a specific capability. Model Studio supports official plugins, third-party plugins, and custom plugins.
Official plugins
The Plug-in Marketplace offers pre-configured official plugins that you can call without setup. See Official plugin instructions.
|
Plugin name |
Tool ID |
Description |
Billing |
|
code_interpreter |
Enables the model to run Python code for tasks like mathematical calculations, data analysis, visualization, and data processing. |
Free |
|
|
calculator |
Lets the model perform complex mathematical calculations, such as "12313 * 13232". |
Free |
|
|
text_to_image |
Lets the model generate images from text, such as "Draw a smiling dog." |
Free for a limited time. Activation is required. |
|
|
quark_search |
Lets the model search online for real-time information and public knowledge, such as "What's the weather in Hangzhou today?" The Quark Search plugin currently retrieves webpage titles, keywords, and summaries, but does not support accessing full webpage content. |
Free for a limited time. Activation is required. |
|
|
generate_qrcode |
Lets the model generate a QR code from a website URL, such as "Generate a QR code for the Model Studio introduction document: https://help.aliyun.com/zh/model-studio/getting-started/what-is-model-studio". |
Free |
|
|
github_search |
Lets the model search for projects on GitHub, for example, "GitHub Search: Qianwen". |
Free |
Third-party plugins
Third-party plugins are also available, covering domains such as business services, image and video, and education. These plugins are tested before listing and require no configuration after enabling. See Third-party plugin instructions.
Custom plugins
You can create custom plugins and integrate them into your application.
Examples
|
Sample input |
Without plugin |
With plugin |
|
What is 12313 × 13232? |
A model cannot reliably solve complex math problems and often returns incorrect answers. The correct answer is 162,925,616. |
With the Calculator plugin, the model performs the calculation accurately. |
Without the plugin, the model attempts the calculation directly and returns an incorrect result that does not match the correct answer of 162,925,616. | When a plugin is available, the model calls the calculator plugin, where the input parameter |
Supported models
|
Model |
Model identifier |
|
Qwen-Turbo |
qwen-turbo |
|
Qwen-Plus |
qwen-plus |
|
Qwen-Max |
qwen-max |
|
Qwen-VL-Max |
qwen-vl-max |
|
Qwen-VL-Plus |
qwen-vl-plus |
Plugin compatibility varies by model. Check the console for details.
See Select a model.
How plugins work
Calling a plugin means calling a tool within it. You can call plugins through an agent application or a workflow application, or the Assistant API.
When a plugin is called from an agent application or the Assistant API, the model determines whether to call a tool based on the user input, tool name, and tool description.
-
If a tool is required, the model selects the appropriate one. The application then calls the tool, combines its output with the original user input, and sends the result back to the model to generate the final response.
-
If no tool is required, the model generates the final response directly.
In a workflow application, calling a plugin means using it as a node in the workflow. The tool is executed as part of a user-defined sequence, rather than the model selecting and calling it autonomously.