Has this been supported or requested before?

• I have checked the GitHub README.
• I have checked the Qwen documentation.
• I have checked the documentation of the related framework.
• I have searched the issues and there is not a similar one.

What is this feature about?

Make the Search function of chat.qwen.ai a callable tool

Proposal

Overview

Currently, the web search functionality in chat.qwen.ai operates in a passive manner — it directly searches for the user’s query without allowing the model to shape or refine the search terms. To improve accuracy and relevance, we propose making the web search function a callable tool that Qwen3 can actively use and control during its reasoning process.

This enhancement would allow Qwen3 to generate optimized search queries based on context and prior conversation history, leading to more accurate and contextually relevant responses.

Problem Statement

The current implementation of the web search function leads to suboptimal results because:

• The model does not have direct control over the search query.
• Context from previous interactions is often lost when performing a search.
• This results in incorrect or misleading answers. For example:
  • User: "What is the capital of China?”
  • Model: "Beijing" (correct).
  • User: "What is the tallest building there?”
  • Model: "Burj Khalifa" (incorrect), as the search was not contextualized to “Beijing".
• This limitation hinders the ability of the model to provide precise, up-to-date information and negatively impacts user experience.

Proposed Solution

Integrate a search function/tool that Qwen3 can explicitly call as part of its response generation pipeline. This will enable the model to:

• Formulate specific and context-aware search queries.
• Use conversational history to adjust and refine search terms.
• Retrieve and synthesize accurate and relevant information from the web.

This approach mirrors the implementations used by other leading models such as ChatGPT, DeepSeek, and ByteDance’s Doubao, where the model has agency over when and how to perform a web search.

引言

目前，chat.qwen.ai 中的网页搜索功能并不理想。它只是简单地对用户的查询进行搜索。相反，模型 Qwen3 应该能够使用一个工具/功能来执行网页搜索。

原因

实现此功能将有助于以下使用场景：

例如当我问它“中国的首都是哪里 ”时，它根据网络搜索结果回答“北京”。然后我再问它“那里最高的建筑是什么 ”，它可能会回答“哈利法塔（Burj Khalifa） ”（世界最高建筑），因为它只搜索了“What is the tallest building there ”这句话，而不是理解为“北京最高的建筑是什么 ”。
如果我们给予模型引导搜索查询的能力，那么搜索结果将会更加准确。这也是 ChatGPT、DeepSeek、字节跳动等公司的搜索功能所采用的方式。

对于依赖 Qwen 获取互联网最新信息的用户来说，这一改进将非常有帮助。
由Qwen3从英文翻译而来

Contributions are welcomed

• I am willing to help implement this feature.