Short answer: yes, if you want frontier-class answers without paying. Qwen Studio puts Qwen3.8-Max, a 2.4-trillion-parameter model, in the free tier with a 1M-token context window, and bundles Deep Research, Thinking mode, image generation, text-to-video, slides and a web-dev mode on top. There is no paid consumer plan to upsell you to; Alibaba makes its money on the API and cloud. The trade-offs are real: Qwen3.8-Max is slow and verbose compared with GPT-6 or Claude, the app filters politically sensitive topics about China, your de-identified chats can be used for training with no in-app opt-out, the iOS app is not listed in the US or most of western Europe, and with Qwen 4 expected imminently, the model line-up on this page may change within weeks.
What is Tongyi Qwen?
Qwen is Alibaba Cloud's family of large language and multimodal models, and Tongyi Qwen / Qwen Studio is the consumer assistant built on top of them. Alibaba launched it as Tongyi Qianwen in April 2023, and it has since become one of the three most-used AI assistants in the world by monthly users, with Alibaba-sourced figures putting the app well past 150 million monthly users by spring 2026 after a surge that started with Qwen3.
What sets it apart from ChatGPT, Gemini and Claude is the business model. The consumer app is completely free, including the flagship model, and there is no Plus, Pro or Ultra tier. Alibaba uses the app to drive adoption and sells access to the same models through its Model Studio API, its cloud, and enterprise tools like DingTalk. The second difference is openness: most Qwen models ship as Apache 2.0 open weights, which is why Qwen is the most-downloaded model family on Hugging Face and the base for thousands of fine-tunes. The largest models (the 2.4T Qwen3.8-Max class) carry a revenue-sharing clause for providers earning over $50 million a year, but everything below that is free to run yourself.
The model line-up, decoded
Alibaba shipped a new Qwen3.x generation roughly every six weeks through 2026, so the names in the app change often. This is what free users get as of early November 2026.
| Model | Released | Type | What it's for in Qwen Studio |
|---|---|---|---|
| Qwen3.8-Max FLAGSHIP | 3 Aug 2026 | Proprietary, 2.4T MoE | The default for hard questions, long documents and coding. 1M-token context. Scores 45.4 on the Artificial Analysis Intelligence Index, close to the frontier, but is slow (~42 tokens/s) and generates far more reasoning tokens than rivals. |
| Qwen3.7-Plus | Jun 2026 | Proprietary | Faster everyday model for chat, drafting and summaries when you don't need Max. |
| Qwen3.8-Omni-Flash | Aug 2026 | Proprietary, omni-modal | Real-time voice, vision and audio understanding; powers the voice chat and camera features in the app. |
| Qwen3.8-Flash-Next | 26 Aug 2026 | Open weights | An experimental preview of the Qwen 4 architecture, released to the open-source community rather than the app. |
| Qwen3.8-27B | Aug 2026 | Open weights, Apache 2.0 | The current dense model to self-host; scores 33.7 on the same index, strong for its size. |
| Qwen-Image 3.0 / 2.1-Turbo | Jul / Oct 2026 | Image | Image generation and editing in the app, with multilingual in-image text and very long prompts. Qwen-Image 3.0 shipped without open weights, unlike earlier versions. |
What's new in 2026 - and what's about to change
Qwen3.8-Max August 2026
Previewed in July as a 2.4-trillion-parameter model and released on the cloud on 3 August, Qwen3.8-Max is the first Qwen flagship to land in the free app with a 1M-token window. Alibaba said it would release the weights too; as of November they are still proprietary.
Qwen 4 is "coming next week" status: not released
Alibaba confirmed on 22 September that Qwen 4 is in training, named four tiers (Max, Plus, Flash, 27B) at its Apsara conference, and its cloud CTO reportedly said "next week" from the Amsterdam stage on 29 October. As of 5 November there are no weights, model card, API name or price. Expect the app's default model to change the day it ships.
Qwen Studio rename and creative tools
The international app dropped the Qwen Chat name in 2026 and absorbed image generation and editing, text-to-video, a slides generator and a web-dev mode. Alibaba also opened the app to third-party services in June, turning it into something closer to a platform than a chatbot.
Leadership change
Junyang Lin, the technical lead behind Qwen since its inception, left Alibaba in March 2026 and founded a Tencent-backed lab. The release cadence has not slowed, but it is worth knowing that the team that built Qwen 2.5 and Qwen 3 is not entirely the team building Qwen 4.
How to use Tongyi Qwen, step by step
1Open the web app or install the desktop app
Go to chat.qwen.ai and sign in with Google, GitHub or email. Mac and Windows apps download from qwen.ai. On iOS, "Qwen Studio" is listed in the UK, Canada, Australia, India, Japan, Singapore and a handful of other stores but not the US or most of western Europe, so the web app is the reliable route. The Android app is on Google Play.
2Pick a model and a mode
Qwen3.8-Max is the default. Toggle Thinking for maths, logic and multi-step coding, Search for anything time-sensitive, and Deep Research when you want a cited multi-page report rather than a chat answer. Switch to Qwen3.7-Plus when speed matters more than depth. Upload PDFs, spreadsheets and images directly into the conversation.
3Create, then export
The Image, Video, Slides and Web Dev entry points live in the sidebar. Generate an image and edit it with follow-up prompts, turn a prompt or an uploaded image into a short clip, draft a deck, or describe a landing page and get working HTML. Everything downloads for free, with no watermark on images we tested. Alibaba does not publish daily limits, but heavy video use does hit queues.
Features that actually matter
Flagship model, free
Qwen3.8-Max is in the free tier. ChatGPT Free runs a smaller model and Claude's free tier is rate-limited; nothing else gives away its best model.
1M-token context
Drop in whole codebases, long contracts or a stack of PDFs. Most consumer free tiers top out at a fraction of this.
Deep Research
A free agentic research mode that searches, reads and compiles a cited report. Paid-tier only on most competitors.
Thinking mode
Extended reasoning for maths, logic and planning. Strong results, but expect long waits: Max is verbose by design.
Image generation & editing
Qwen-Image handles very long prompts and renders text in 10+ languages, including Chinese, better than most Western models.
Text and image to video
Short clips from a prompt or a still. Useful for social drafts; not a Kling or Seedance replacement, and the queue can stall.
Omni voice and vision
Qwen3.8-Omni-Flash powers real-time voice conversation and camera understanding in the mobile apps.
Multilingual strength
Trained on 100+ languages with notably good Chinese, Arabic, Japanese and south-east Asian coverage.
Open weights to self-host
Don't trust the app with your data? Run Qwen3.8-27B locally under Apache 2.0 and keep everything on your machine.
Tongyi Qwen pricing (2026)
For consumers, the pricing section is short: it's free. There is no paid tier to compare, which is unusual enough that several third-party "Qwen Pro" pricing pages you'll find in search are either describing the developer API plans or are unofficial resellers. Here is how the real options line up.
| Product | Price | What you get | Who it's for |
|---|---|---|---|
| Qwen Studio app FREE | $0 | Qwen3.8-Max, 3.7-Plus and Omni-Flash ยท Thinking ยท Search ยท Deep Research ยท image, video, slides, web dev ยท 1M context | Everyone. No published limits; no paid upgrade exists |
| API pay-as-you-go (Model Studio) | From $0.05 / 1M tokens | Qwen3.8-Max at $2.00 in / $6.00 out per 1M tokens, flat across the 1M window; cheaper Plus, Flash and Turbo tiers. New accounts get ~1M free tokens per model for 90 days | Developers and businesses |
| Token Plan (Personal) | $6 / $10 / $18 / $68 per mo (promo; list $8โ$80) | 11,500 to 180,000 credits a month for API and agent tools; Team seats from $20 | Individual builders who want predictable API billing |
| Coding Plan Pro | $50 / mo | Up to 90,000 model calls a month (6,000 per rolling 5 hours) in Qwen Code and supported IDE tools | Heavy agentic-coding users; Lite tier closed to new buyers |
| Self-hosted open weights | $0 + hardware | Qwen3.8-27B and smaller under Apache 2.0; a 27B model needs roughly a 24 GB GPU at 4-bit | Privacy-first users and teams |
Prices checked against Alibaba Cloud Model Studio and qwen.ai in early November 2026; promotional Token Plan prices are marked "limited time" and may revert to list.
How good are the models, really?
Independent benchmarks put Qwen3.8-Max just below the top tier and ahead of every other free offering. On Artificial Analysis' Intelligence Index it scores 45.4, which lands it near Gemini 3.1 Pro and the GPT-6 mid-tier rather than the absolute frontier (Claude Fable 5 and GPT-6 Sol sit higher). Its weakest spots in that evaluation are agentic terminal work (38.9% on Terminal-Bench 4.0) and factual recall (31.7% on AA-Omniscience), and its most striking trait is verbosity: it burned roughly 190 million tokens to complete the index against a median of 85 million, at 42 tokens per second. In plain terms, it thinks hard and slowly.
In day-to-day use that translates to excellent long-document work, strong maths and code, very good multilingual output, and a noticeable wait on anything with Thinking enabled. Coding is where the community is warmest: Qwen3-series models have sat in the global top three for agentic coding in community leaderboards through 2026, and Qwen Code with the Coding Plan is a credible free-to-cheap alternative to Claude Code or Codex for solo developers. The repetition and language-drift problems that dogged early Qwen3 are largely gone in 3.7 and 3.8.
What users say
The free flagship and the bundled creative tools earn Qwen a lot of goodwill: reviewers consistently score it 4.0 to 4.5 and call it the best value in the category. The complaints cluster in four places. Speed: Max is slow and Thinking mode can take minutes. Video: Google Play reviews describe generations stalling and a one-video-per-image limitation, and some users say the ads promise more than the app delivers. Content filtering: an NDSS 2026 study from Michigan and Berkeley found the earlier Qwen2.5-Max app blocked roughly a fifth to a third of politically sensitive queries about China, mostly before the model even answered; no equivalent test of the current app has been published, but reviewers still report the behaviour. Availability: the iOS app is missing from the US and most European stores, and some corporate networks block Alibaba domains. Users in Asia, the Middle East and Latin America are markedly happier than those in the US and EU.
Pros and cons
What we liked
- Flagship Qwen3.8-Max in the free tier, with no paid plan above it
- 1M-token context window at no cost
- Deep Research, Thinking mode and web search all free
- Image generation, editing, video, slides and web-dev mode in one app
- Top-three agentic coding; Qwen Code plus Coding Plan is a bargain for developers
- Best-in-class Chinese and strong multilingual output, including in-image text
- Apache 2.0 open weights for most models, so you can self-host
- Rapid release cadence; Qwen 4 is imminent
What held it back
- Qwen3.8-Max is slow and extremely verbose compared with GPT-6 and Claude
- De-identified chats may be used for training with no in-app opt-out
- Filters politically sensitive topics about China
- iOS app not listed in the US or most of western Europe
- Video generation is flaky and limited; not a Kling replacement
- No published usage limits, so heavy users can't plan around them
- Factual recall and terminal agent benchmarks trail the leaders
- Model names change every few weeks; Qwen 4 will reshuffle the line-up again
Who should use Tongyi Qwen - and who shouldn't
Worth it for
- Students, researchers and writers on a budget who want a frontier-class model, Deep Research and a huge context window without a $20 subscription.
- Developers who want to try an open model family in the app and then self-host or use the cheap API; the Coding Plan is one of the best deals in agentic coding.
- Multilingual users, especially anyone working in Chinese, Japanese, Arabic or south-east Asian languages, where Qwen often beats Western models.
- Creators who want everything in one tab: image, video, slides and a landing page from the same chat, free.
- Businesses already on Alibaba Cloud or DingTalk, where the integration is native.
Not worth it for
- Anyone handling confidential or regulated data in the consumer app; use the API or self-host instead.
- Users who need fast answers; for quick back-and-forth, Gemini Flash or GPT-6 Luna feel twice as responsive.
- Journalists and researchers covering Chinese politics, where filtering will get in the way.
- US and EU iPhone users who want a native app; the web app works, but the App Store listing isn't there.
Tongyi Qwen vs the alternatives
| Assistant | Free tier model | Paid from | Strength | Weakness vs Qwen |
|---|---|---|---|---|
| Tongyi Qwen / Qwen Studio | Qwen3.8-Max (flagship) | No consumer tier | Flagship free, 1M context, Deep Research, creative tools, open weights | Slow, China filtering, training on chats, patchy iOS availability |
| ChatGPT (GPT-6) | GPT-6 Luna | $8 Go / $20 Plus | Fastest ecosystem, best agents and voice, no-training toggle | Flagship Sol is paywalled; free context is small |
| Claude | Sonnet 5.5 (limited) | $20 Pro | Strongest writing and coding; best privacy defaults | Free tier is tightly rate-limited; no image or video generation |
| Gemini | 3.6 Flash | $19.99 AI Pro | Speed, Google Workspace integration, Veo video on paid tiers | 3.1 Pro and 2M context need a subscription |
| DeepSeek | V4-Flash (V4-Pro preview) | Free | Free, open weights, strong reasoning, ~1M context | Bare-bones app: no image, video or slides; same filtering and data concerns |
| Kimi (Moonshot) | K2.7 | Paid tiers for K3 / Turbo | Excellent long-horizon coding (K3); OK Computer agent; slides | Best model paywalled; smaller ecosystem |
Qwen or DeepSeek?
Both are free, Chinese, open-weight and strong on reasoning, so this is the comparison most readers actually want. Choose Qwen for the complete app: Deep Research, images, video, slides, web dev, voice and a 1M-token window in one place, plus a far bigger ecosystem of fine-tunes and tooling. Choose DeepSeek if you only want a chat window with a top reasoning model and prefer its terser, faster responses. On raw coding quality the two have traded places all year and are close to a toss-up; if you're choosing for coding specifically, test both on your own repo. Neither is the right place for confidential work; for that, pick Claude or ChatGPT with training turned off, or self-host either family.
Final verdict: is Tongyi Qwen worth it?
Yes. For the price of nothing, Tongyi Qwen is the most capable AI assistant available, and the gap between its free tier and everyone else's is not close. A 2.4T flagship with a 1M-token context, a free Deep Research agent, and a full creative toolkit is a package that costs $20 to $40 a month elsewhere. For students, multilingual users, developers and anyone who simply doesn't want another subscription, it should be the first thing you try.
What stops it from being the only assistant you need is the small print and the stopwatch. Qwen3.8-Max is slow and wordy, the app trains on your de-identified chats without an opt-out, politically sensitive topics about China are filtered, and the iOS app isn't in the US store. Video is a nice extra, not a reason to switch. And the model line-up is about to be reshuffled by Qwen 4, which Alibaba has been promising "next week" since late October.
Our advice: use Qwen Studio as your free workhorse for research, long documents, coding help and creative drafts; keep ChatGPT or Claude for anything confidential or where speed matters; and check back the week Qwen 4 lands, because the free tier is about to get better again.
Try it with a real task
Open chat.qwen.ai, upload the longest PDF you have, ask a Thinking-mode question about it, then run one Deep Research report. That's a $20 subscription's worth of features in ten minutes, and it costs nothing.
Visit Qwen โFrequently asked questions
Is Tongyi Qwen free?
Yes. The consumer app, Qwen Studio at chat.qwen.ai and the mobile and desktop apps, is free with no subscription, and it includes the flagship Qwen3.8-Max model, Thinking mode, web search, Deep Research, image generation, video, slides and a web-dev mode. Alibaba does not publish usage limits. Paid plans exist only for the developer API: pay-as-you-go tokens, Token Plans from $6 a month, and a $50 Coding Plan.
What is the difference between Tongyi, Qwen and Qwen Studio?
Tongyi Qianwen (้ไนๅ้ฎ) is the Chinese name and the domestic app at tongyi.aliyun.com. Qwen is the model family and the brand developers use. Qwen Studio, formerly Qwen Chat, is the international consumer app at chat.qwen.ai, operated by NTH Power Global Tech Singapore. They run the same models.
Is Qwen 4 out yet?
Not as of 5 November 2026. Alibaba confirmed on 22 September that Qwen 4 is in training and named four tiers: Qwen4-Max, Plus, Flash and 27B. Qwen3.8-Flash-Next, released in August, is an open-weight preview of the Qwen 4 architecture. Alibaba Cloud's CTO reportedly said "next week" on 29 October, but no weights, model card, API name or price have appeared. When it ships, expect it to replace Qwen3.8-Max as the app's default.
Is Qwen safe to use? What happens to my data?
The international app is operated from Singapore and the privacy policy dated 7 September 2026 says personal data is stored in Singapore and processed by Alibaba Cloud Singapore. The operator may use de-identified chats and feedback to train models and there is no in-app opt-out. EU and UK transfers rely on adequacy decisions and standard contractual clauses. For confidential work, use the API, which has separate commercial terms, or self-host the open-weight models.
Does Qwen censor content?
The app filters politically sensitive topics about China. A University of Michigan and UC Berkeley study presented at NDSS 2026 found the earlier Qwen2.5-Max app blocked 16 of 80 English and 24 of 80 Simplified Chinese sensitive queries at least once, mostly before the model answered. No comparable test of the current Qwen Studio has been published, but reviewers continue to report the behaviour. General-purpose, technical and creative use is unaffected.
Is Qwen better than ChatGPT?
On free tiers, Qwen wins: it offers its flagship Qwen3.8-Max with a 1M-token window, while ChatGPT Free runs the smaller GPT-6 Luna. Against ChatGPT Plus or Pro, GPT-6 Sol is stronger, faster and backed by a bigger agent and voice ecosystem, with a no-training privacy toggle. If you won't pay, pick Qwen; if you will, ChatGPT or Claude still lead.
Can I run Qwen on my own computer?
Yes. Most Qwen models are released as open weights under Apache 2.0 on Hugging Face and ModelScope. Qwen3.8-27B is the current dense model to self-host and runs on a single 24 GB GPU at 4-bit quantisation; smaller variants run on laptops. The largest models, such as the 2.4T Qwen3.8-Max, remain proprietary, and a revenue-sharing clause applies to providers earning over $50 million a year on the biggest open models.
Is the Qwen iPhone app available in the US?
As of October 2026, the Qwen Studio iOS app is listed in the UK, Canada, Australia, New Zealand, India, Brazil, Mexico, Japan, Singapore and the UAE, but not in the US, Germany, France, Spain, Italy or several other markets. The web app at chat.qwen.ai, the Android app and the Mac and Windows desktop apps work everywhere.
Model availability, pricing, privacy terms and Qwen 4 status verified against qwen.ai, Alibaba Cloud Model Studio, the Qwen Hugging Face organisation and Alibaba's Apsara Conference announcements in early November 2026, and cross-checked against Artificial Analysis, Fello AI, AI Miracle, Google Play reviews and the NDSS 2026 chatbot-filtering study. Alibaba ships new Qwen models every few weeks, and Qwen 4 is expected imminently; confirm the current line-up in the app before relying on this page. This review may contain affiliate links; it does not affect our scoring.