Qwen updates DeepResearch agent to version 2511
Alibaba's Qwen team has released DeepResearch 2511, which adds a slower Advanced Mode, file uploads and a reworked multi-agent design, and says it improves on version 2507 across every measure it tested.
Alibaba's Qwen team announced Qwen DeepResearch 2511, an update to its research agent. The team says the release improves on the previous version, 2507, "comprehensively and significantly." It measured anti-hallucination rate, report comprehensiveness, reliability, instruction following, report length, search depth and research time. The post doesn't include the benchmark figures, name the test sets or compare the tool with other research agents.
The update adds two modes. Normal Mode is for quick, everyday queries. Advanced Mode "invests more time and computing power" for deeper multi-step reasoning. Users can now upload PDFs, Excel spreadsheets and images, and the system works their contents into its searches and reasoning.
Qwen says it rebuilt how the system searches and reads the web, so it can process more high-quality sources in a given time. The team says this reduces hallucinations and repetition. Users can set a report's word count, paragraph structure and level of detail. Findings can also be turned into a web page or a podcast with one click.
On the technical side, the team describes a multi-agent system that is optimised jointly across several stages:
- analysing what the user wants
- planning the research
- verifying search results
- running tool calls in parallel
- reading web pages
- writing the report
Memory management and research-state tracking tie these stages together.
The announcement doesn't cover pricing, availability by region, API access or which underlying models power the system.