Qwen DeepResearch 2511 Release

Qwen DeepResearch 2511 is an AI-powered research assistant designed to minimize the friction between initial curiosity and deep insight by automating the heavy lifting of information retrieval, verification, and synthesis. By reducing the perceived cost of exploration, the system aims to make deep research a natural extension of the thinking process rather than a high-barrier task.

Key Features and Capabilities

Qwen DeepResearch 2511 introduces several qualitative improvements in research depth and efficiency:

Dual-Mode Operation

  • Normal Mode: Optimized for high-efficiency, general-purpose use to meet daily research needs quickly.
  • Advanced Mode: Allocates additional computing power and time to perform comprehensive analysis and deeper multi-step reasoning.

Data Integration and Retrieval

  • Local File Integration: The system supports the upload and analysis of various file formats, including PDF, Excel, and images, integrating this content directly into the reasoning process.
  • Enhanced Search and Reading: Reconstructed web search and comprehension strategies are used to refine high-quality information sources, which reduces redundancy and hallucinations while increasing the reliability of results.

Output and Interaction

  • Controllable Reporting: Report writing is optimized to integrate user intent, allowing for flexible control over detail levels, paragraph structure, and word count.
  • Diverse Output Formats: Beyond traditional reports, the system can generate web pages and podcasts with a single click.
  • Interactive Experience: The interface features a clearer information hierarchy and intuitive citation marking for a more immersive experience.

Technical Architecture

Qwen DeepResearch 2511 is powered by a multi-agent collaborative mechanism. This architecture involves joint optimization across several critical research stages:

  • User Need Analysis: Understanding the core intent of the research query.
  • Strategy Planning: Developing an overall plan for the research task.
  • search Verification: Implementing strategies to verify the information found.
  • Parallel Tool Calls: Executing multiple tool calls simultaneously to increase efficiency.
  • Web Page Analysis: reading and analyzing content from the web.
  • Report Writing: Synthesizing findings into a final document.

This multi-agent approach is supported by an overall memory management mechanism and research state tracking to maintain coherence across complex tasks.

Performance Improvements

In systematic evaluations, Qwen DeepResearch 2511 demonstrates significant improvements over version 2507 across the following dimensions:

  • Anti-hallucination rate
  • Report comprehensiveness and reliability
  • Instruction following
  • Report length
  • Search depth
  • Research time

Sources