Qwen Research: Reports, Webpages & Podcasts

What is Qwen Research?

Qwen Research officially called Qwen Deep Research is Alibaba's AI-powered research assistant feature inside the Qwen Chat web app. Where a normal LLM gives you a quick answer, Qwen Deep Research goes much further: it asks clarifying questions to scope your request, plans a multi-step research workflow, autonomously searches and reads dozens of sources from the open web, synthesizes the findings into a structured report with citations, and uniquely can turn that report into an interactive webpage and a multi-speaker podcast with one or two extra clicks. The whole loop, from "I have a question" to "I have a multimedia deliverable," runs entirely inside the chat interface.

The current generation, called DeepResearch 2511 (released November 2025), positions Qwen as a direct competitor to OpenAI's ChatGPT Deep Research, Google's Gemini Deep Research, and Anthropic's research tools. What sets the Qwen version apart isn't the report quality alone it's the integrated multimodal output. After ChatGPT or Gemini produce a research report, you're done; with Qwen, that's actually step one of three. The same research can become a live webpage you can share publicly and a podcast you can listen to on a walk, all without leaving Qwen Chat.

Under the hood, Qwen Deep Research is a coordinated pipeline of several Qwen models working together: Qwen3-Coder generates the webpage structure and HTML/CSS, Qwen-Image creates inline graphics and visualizations, and Qwen3-TTS powers the dynamic two-host podcast narration. While the underlying models are open-source, the orchestrated Deep Research workflow itself is a proprietary hosted service available through Qwen Chat and the DashScope API. You don't need to configure anything the entire stack runs in Alibaba's cloud.

Key Features

Qwen Deep Research bundles a handful of capabilities that, taken together, replace what used to require a research assistant, a web designer, and a podcast producer. The headline features:

• Cited research reports: Comprehensive multi-section reports with inline citations linking to original sources, so claims are auditable rather than just plausible-sounding.
• Clarifying questions: Before diving in, Qwen asks follow-up questions to scope your request history vs theory vs applications, depth, audience, length so the output matches what you actually need.
• Web search and synthesis: Autonomously runs dozens of search queries, fetches and reads top results, identifies data conflicts between sources, and synthesizes coherent findings.
• Interactive webpages: One click to convert any report into a live, publicly shareable webpage with proper layout, navigation, embedded images, and responsive design.
• Multi-speaker podcasts: Two AI hosts in genuine back-and-forth conversation (not a monologue read-aloud) covering the report's key points, with multiple voice options to choose from.
• Dual mode (Normal + Deep): Normal Mode for quick everyday research, Deep Mode for long-form rigorous reports. You pick the depth that matches the task.
• File uploads: Feed Qwen Deep Research your own PDFs, research papers, spreadsheets, or images alongside its web searches. Reports incorporate your materials and external sources together.
• Data analysis built in: Calculates compound growth rates, identifies data discrepancies between sources, and generates visual reports with charts and figures embedded inline.
• Free for everyone: Available globally in the free Qwen Chat web app no Pro subscription required for basic Deep Research access.

How Qwen Research Works

The workflow is deliberately conversational, not configuration-heavy. Here's what actually happens when you ask Qwen Chat to do research on a topic.

Step 1 You ask, Qwen scopes

You start by typing a topic in Qwen Chat with Deep Research mode enabled. Rather than diving straight into a generic answer, Qwen asks a clarifying question first. For a query like "tell me about quantum physics," it might ask whether you want the history, the theoretical foundations, current research frontiers, or real-world applications. If you're not sure, you can simply tell it to surprise you and Qwen picks an angle. This scoping step is what separates a deep research workflow from a regular chat the model is building a research plan before it commits to writing.

Step 2 Qwen plans and executes

Once it has scope, Qwen builds a structured research plan typically 5–15 subtopics or questions it intends to investigate and then executes that plan autonomously. It runs search queries, fetches and reads the top web pages for each, extracts relevant facts and quotes, cross-references claims between sources, and surfaces data conflicts when sources disagree. This phase is where Deep Research earns its name: a single query can trigger dozens of web fetches and many minutes of model reasoning. Qwen shows its progress streaming in real time, so you can watch the searches and intermediate notes as they happen rather than staring at a blank loading screen.

Step 3 Report generation

With all the research done, Qwen synthesizes the findings into a structured report. Expect a clear executive summary, multi-section body with subheadings, embedded data tables and charts where appropriate, and inline citations linking back to the original sources. Reports typically run 2,000–8,000 words depending on complexity and the depth setting you chose. You can chat back to refine the report, ask follow-up questions, or request specific sections be expanded.

Step 4 One-click webpage

When the report is ready, you can click Create webpage and Qwen3-Coder generates a complete, styled HTML page hosted on Qwen's infrastructure. The page includes proper typography, navigation between sections, embedded Qwen-Image graphics, and a shareable public URL. This is essentially "publish to web" without WordPress, SEO setup, or any HTML knowledge.

Step 5 One-click podcast

The final step is the podcast. Click Create podcast and Qwen3-TTS produces a two-host audio episode discussing the report's findings not a robotic read-aloud, but an actual back-and-forth conversation between two AI hosts with distinct voices. You can choose between several voice combinations, including different accents and tones. Episode length is typically 5–15 minutes depending on the report's depth.

The Models Behind Qwen Research

Qwen Deep Research isn't a single model it's an orchestrated pipeline of specialized Qwen models, each handling the part of the workflow it's best at. Understanding the architecture helps explain why the output is so coherent across formats.

The research and reasoning phase is driven by the latest Qwen3 reasoning models (Qwen3-Max or Qwen3-Plus depending on configuration), which handle query planning, web reading, source synthesis, and the actual report writing. These are the same models exposed in the public Qwen API, but here they're configured with web search tools, document retrieval, and a multi-step agent loop that lets them work autonomously for minutes at a time.

The webpage generation phase uses Qwen3-Coder, the same code-specialized model that powers Qwen's coding products. It takes the structured report and generates clean, responsive HTML, CSS, and JavaScript for the published page, including layout choices appropriate for the report's content a market analysis gets data-heavy formatting, a historical overview gets a timeline structure, a how-to guide gets numbered steps with visual aids.

Inline graphics and visualizations are produced by Qwen-Image, Alibaba's image generation model, which creates contextually relevant figures based on the report's content. Finally, the podcast generation phase is handled by Qwen3-TTS, released in January 2026, which supports voice cloning, voice design, and natural multi-speaker dialogue across 10+ languages. The two-host conversational format is what gives Qwen's podcasts their dynamic feel compared to single-narrator AI audio.

How Qwen Research Compares

Deep research as a feature category emerged in 2024–2025, and there are now four serious contenders. Each has distinct strengths.

Qwen Deep Research is the integration champion it's the only major option that produces reports, webpages, and podcasts in a single workflow without leaving the chat. It's also free, available globally, and notably strong at balancing multiple perspectives on contested topics. The trade-off is that the podcast voice quality is inconsistent some voices sound natural, but others produce robotic delivery or awkward accents. Google's NotebookLM and Gemini still produce noticeably more human-sounding podcasts.

OpenAI ChatGPT Deep Research tends to write the most polished, pedagogically clear reports well-structured, comprehensive, and reader-friendly. It's tightly integrated with the ChatGPT ecosystem and works well for academic-style writing. Drawback: it's gated behind a paid subscription and doesn't produce webpages or podcasts natively.

Google Gemini Deep Research excels at integrating scientific frontiers with broader context it's particularly good at topics that span multiple domains (consciousness, evolutionary biology, cosmology). Its Audio Overviews feature, inherited from NotebookLM, produces the most natural-sounding podcasts in the space. Drawback: also gated behind paid Gemini Advanced.

Anthropic Claude with research tools is the strongest option for nuanced analysis on contested or subjective topics strong reasoning, careful caveats, balanced framing. Drawback: doesn't have built-in autonomous web research at the same scale, and no native podcast or webpage output.

For most users, the choice reduces to: ChatGPT or Gemini if you want polished single-format reports and don't mind paying; Qwen if you want the full report-plus-webpage-plus-podcast pipeline for free; Claude if your priority is careful reasoning on hard subjective questions.

Use Cases

Qwen Deep Research's combination of cited reports plus multimedia output makes it a strong fit for several distinct workflows.

Market research and competitive analysis is the most common professional use case feed Qwen a question about an industry segment or competitor, get back a structured report with growth rates, market sizing, and key player analysis, then convert that into a webpage to share with your team or a podcast to listen to on your commute. The Alibaba launch demo specifically showcased this with U.S. SaaS market analysis.

Investment and due diligence work benefits from the citation-heavy output analysts can quickly survey a company, sector, or technology trend with all sources linked back to originals for verification. The data analysis features (compound growth calculations, discrepancy detection) reduce the manual spreadsheet work that usually surrounds this kind of research.

Academic and educational research uses Deep Research for literature surveys, topic exploration before a paper, or generating teaching materials. Students get the cited report as a starting point; teachers get a webpage and a podcast they can hand to students as learning resources on the same topic.

Content creation teams use it to research a topic, publish the resulting webpage as a long-form article, and ship the podcast as an episode all from one initial prompt. This dramatically compresses the timeline from idea to multi-channel publication.

Personal learning and curiosity rounds out the list anyone who wants to deeply understand a topic, from quantum mechanics to a country's history to a programming concept, can get a cited report and listen to the podcast version while doing something else. The "tell me about X like I'm preparing for a smart dinner conversation" workflow is genuinely well-served here.

Qwen Research API

Qwen Deep Research is also available programmatically through the Alibaba Cloud Model Studio API, exposed via the qwen-deep-research family of models. The current generation is qwen-deep-research-2025-12-15, which exposes a streaming interface that emits progress events as the research runs.

The API surfaces three streaming event types that let your application show real-time progress to users:

• streamingThinking emitted as the model decomposes the research task and summarizes web content (the unified event type in the December 2025 release).
• streamingQueries emitted as the model generates and runs search queries (used in the original qwen-deep-research model).
• streamingWebResult emitted as the model performs web searches and analyzes returned content (used in the original qwen-deep-research model).

A minimal Python example using the DashScope SDK:

import os
import dashscope

dashscope.api_key = os.getenv("DASHSCOPE_API_KEY")
dashscope.base_http_api_url = 'https://dashscope-intl.aliyuncs.com/api/v1'

response = dashscope.Generation.call(
    model="qwen-deep-research-2025-12-15",
    messages=[{
        "role": "user",
        "content": "Research the current state of quantum computing "
                   "hardware and major commercial players."
    }],
    stream=True,
)

for chunk in response:
    event_type = chunk.output.get("event_type")
    if event_type == "streamingThinking":
        print(f"[thinking] {chunk.output.thinking}")
    elif event_type == "result":
        print(f"\n[REPORT]\n{chunk.output.text}")

The API is well suited to integrating Deep Research into your own applications a customer dashboard, an internal knowledge tool, a research SaaS where you want the research engine but your own UI around it. For full documentation see Build Research Reports with Qwen Deep Research API.

How to Use Qwen Research

Getting started with Qwen Research takes about a minute. Here's the exact path:

1. Open Qwen Chat at chat.qwen.ai in any modern browser, or download the Qwen app for iOS or Android.
2. Sign in with an Alibaba Cloud, Google, or phone-number account. Free signup is fine no Pro plan required for basic Deep Research access.
3. Start a new chat and look for the Deep Research toggle in the input area (sometimes shown as a magnifying glass or research icon).
4. Choose Normal or Deep mode. Normal Mode is best for quick everyday queries that need web grounding; Deep Mode is for longer, more rigorous reports.
5. Type your topic and submit. If Qwen asks a clarifying question, answer it or tell Qwen to pick an angle for you.
6. Wait while research runs. Depending on depth, expect anywhere from 30 seconds to 5+ minutes. You'll see streaming progress (searches, found sources, intermediate notes) the entire time.
7. Review the report when it's done. Click through citations to verify claims. You can chat back to refine sections, expand topics, or request changes.
8. Click "Create webpage" to generate the shareable HTML page, or "Create podcast" to generate the two-host audio version. Both are optional you can use just the report if that's all you need.

Tips for Better Results

A few pragmatic suggestions for getting the most out of Qwen Deep Research.

Be specific in your initial prompt. "Tell me about AI" produces a generic survey; "Compare the agentic coding capabilities of Claude Opus, GPT-5, and Qwen3-Coder-Next as of Q1 2026" produces a focused, useful report. The clarifying questions help, but specificity in the original prompt always wins.

Use file uploads to ground the research. If you have internal documents, a competitor's annual report, or your own data, attach those files. Qwen will incorporate them alongside web sources, which dramatically improves relevance for proprietary topics that public search wouldn't cover well.

Pick Deep Mode for rigor, Normal Mode for speed. Deep Mode reads more sources, asks more clarifying questions, and produces longer reports at the cost of taking several times longer. For quick fact-finding, Normal Mode is plenty.

Treat the podcast voices as configurable. The default voices aren't always the best fit; spend 30 seconds picking voices that match the tone of your topic. A finance market report sounds better with measured professional voices; a pop-culture explainer sounds better with more casual ones.

Verify critical claims. Like any AI research tool, Qwen Deep Research occasionally hallucinates or misattributes facts. For anything you'll publish, present, or make decisions on, click through to the cited sources and verify the load-bearing claims yourself. The citations make this fast that's their point.

FAQ

Is Qwen Research free?

Yes. Qwen Deep Research is available in the free Qwen Chat web app and mobile apps at no cost no Pro subscription required for standard usage. Heavy API usage through Model Studio is billed per token, but the consumer Qwen Chat experience is free worldwide.

How long does a research report take to generate?

Anywhere from 30 seconds for a quick Normal Mode query to 5–10 minutes for a deep, multi-section Deep Mode report. You can grab a coffee while it runs Qwen streams progress the whole time so you'll see when it's done.

Can I upload my own files?

Yes. Since DeepResearch 2511 (November 2025), Qwen Research supports document and image uploads. Attach PDFs, research papers, spreadsheets, or images and Qwen will incorporate them into the analysis alongside its web searches.

Are the cited sources reliable?

Qwen pulls from the open web, so source quality reflects what's available usually a mix of news articles, company sites, Wikipedia, academic abstracts, and industry reports. Citations link directly to originals, so you can verify each claim yourself. For peer-reviewed academic rigor, treat Qwen Research as a starting point, not a final source.

Can I share or publish the webpages it creates?

Yes. Webpages generated by Qwen Deep Research get a public shareable URL hosted on Qwen's infrastructure. You can share the link freely. For long-term hosting on your own domain, you can also copy the underlying HTML and self-host.

How good are the podcasts compared to Google NotebookLM?

Mixed. Qwen's two-host format and large voice selection are real strengths, and many voices sound natural. However, Google NotebookLM and Gemini still produce more consistently human-sounding podcasts with smoother conversational flow. For pure podcast quality, NotebookLM wins; for integrated report-plus-webpage-plus-podcast workflow, Qwen wins.

Is there an API for Qwen Research?

Yes. The qwen-deep-research model family is available through Alibaba Cloud Model Studio's DashScope API. The current model is qwen-deep-research-2025-12-15, with streaming progress events for thinking, queries, and web results.

Does Qwen Research work in languages other than English?

Yes. Qwen models broadly support 100+ languages, with particularly strong performance in English and Chinese. You can ask research questions in your preferred language and Qwen will respond in the same language, citing sources in whatever language they were originally published.

Can Qwen Research access paywalled academic papers?

No. Qwen Research uses the open web, which means it can read abstracts and freely accessible papers but not subscription journal articles. For rigorous academic work, supplement Qwen's reports with your own access to paywalled sources.

What's the difference between Normal Mode and Deep Mode?

Normal Mode is fast and lightweight good for everyday research where you want grounded answers with citations but don't need a full report. Deep Mode runs a longer, multi-step research process with more sources, more clarifying questions, and longer reports. Use Deep Mode for serious research tasks; use Normal Mode for quick web-grounded queries.

Final Thoughts

Qwen Deep Research is one of the most interesting product moves Alibaba has made in the consumer AI space. By bundling cited reports, interactive webpages, and multi-speaker podcasts into a single workflow and offering it for free, globally Qwen has carved out a real category position against ChatGPT, Gemini, and Claude. The individual pieces aren't always best-in-class (Gemini's podcasts sound more human, ChatGPT's reports are more polished), but the integration is genuinely novel, and for many workflows the integration is what actually matters.

The easiest way to evaluate it is to just try it. Open chat.qwen.ai, enable Deep Research mode, and ask it to research a topic you'd otherwise spend an hour reading about. Five minutes later you'll have a report, a shareable webpage, and a podcast and a much better sense of whether Qwen Research belongs in your workflow.