Alibaba

Qwen

Provider

Alibaba

Price

$0 (free app)

Platforms

Web, iOS, Android

Launched

Nov 2025

Updated June 2026

Qwen is Alibaba’s free AI assistant and one of the most-used chatbots in the world. The app — launched as Qwen Chat and renamed Qwen Studio on 10 April 2026 — costs nothing, has no consumer subscription, and runs Alibaba’s Qwen family of models for multilingual, multimodal work across text, images, video and voice. Its current flagship, Qwen3.7-Max, scores 80.4% on SWE-bench Verified on Alibaba’s own testing and ranks in the global top ten on Artificial Analysis’s independent intelligence index. The catch is data: Qwen is built by a China-headquartered company, so for sensitive or regulated work the safer route is self-hosting Qwen’s open weights rather than using the hosted app.

This is the rare frontier-class assistant that is genuinely free, with image generation, video, deep research and slide creation in the box — features rivals charge $20 a month for. This guide covers the free Qwen Studio app across web, mobile and desktop: what the models can do, where the app excels, the privacy trade-offs of a China-based provider, and how it compares to DeepSeek, Kimi and ChatGPT.

Quick stats

Pricing breakdown

Qwen’s pricing is unusual: the consumer app is entirely free, and there is no flat monthly plan in between free and the developer API. You pay only if you build on Qwen through Alibaba Cloud’s API.

Free — $0

The Qwen Studio app at chat.qwen.ai is free with no subscription, unlike ChatGPT Plus or Claude Pro. The free tier includes:

• Qwen’s current models, including the Qwen3.7-Max flagship and the open Qwen 3.6
• Thinking mode — visible chain-of-thought reasoning for maths, logic and coding
• Web search — current information with citations
• Deep Research — a research agent that plans, searches multiple sources and returns a structured report
• Image generation and editing (Qwen-Image), video generation and slide creation
• Document and image understanding with file uploads
• Voice and video chat on mobile
• Web Dev — describe a site and get a working webpage with clean code
• 119+ languages, with no advertised message caps

There is no consumer paid tier. The trade-offs are about data and language quality, not paywalls — covered below.

API — pay-as-you-go (developers)

Building on Qwen is billed per token through Alibaba Cloud Model Studio (the platform formerly branded DashScope). Rates run from $0.05 / $0.40 per million tokens on Qwen-Flash to $1.25 / $3.75 on the flagship Qwen3.7-Max (a 50% launch promotion; list price $2.50 / $7.50), all in USD. The standing free API tier ended on 15 April 2026 and was replaced by a one-time trial of 1 million tokens per model (about 70 million in total) over 90 days on the Singapore endpoint. Full API pricing sits on the Qwen model page; the consumer app needs none of it.

Models powering Qwen

In the app you pick a model from a dropdown. Alibaba runs a deliberate two-tier strategy: open-weight models for the mid-tier and a closed-weight model at the frontier. For full benchmarks and pricing, see the Qwen model page.

Qwen3.7-Max (May 2026)

The current flagship, announced on 20 May 2026 at the Alibaba Cloud Summit and the first Qwen flagship to ship closed-weights only:

• 80.4% on SWE-bench Verified on Alibaba’s own testing — within ~0.2 points of DeepSeek V4 (80.6%) at the top of the open-weight cluster
• Artificial Analysis Intelligence Index ~56.6 — an independent score that places it in the global top ten
• 1M-token context window; reasoning-native, with Alibaba reporting it can run external coding harnesses (including Anthropic’s Claude Code) for long autonomous sessions
• Closed weights — developers cannot download it, but anyone can chat with it free in the app

Alibaba’s head-to-head coding figures (for example Terminal-Bench 2.0 and SWE-bench Pro leads over Claude Opus 4.7) are vendor-run and should be read as a ceiling; the independent Artificial Analysis index is the more conservative read.

Qwen 3.6 (April 2026)

The current open-weight release, shipped under the Apache 2.0 licence in April 2026 (a 35B-A3B Mixture-of-Experts model and a dense 27B variant). It is the model to self-host if you want Qwen’s capability without sending data to Alibaba.

Qwen 3.5 (February 2026)

The previous open flagship: a 397B-parameter Mixture-of-Experts model with about 17B active parameters, a 262K context window extensible to 1M tokens, and native support for 201 languages — also Apache 2.0.

Specialist models

The app routes specific tasks to dedicated models:

• Qwen-Image — image generation and editing at native 2K resolution, with unusually accurate text rendering inside generated images
• Wan — text-to-video and image-to-video generation
• Qwen3-Omni — real-time voice and audio across 119 text and 19 speech-input languages
• Qwen3-VL — vision-language understanding of images and long video
• Qwen3-Coder — Alibaba’s coding model (the 480B-parameter Qwen3-Coder powers agentic coding workflows)

Features deep dive

Deep Research

Qwen’s Deep Research mode works like a research agent. Give it a complex question and it asks clarifying questions to scope the work, then autonomously searches the web across multiple sources, reconciles discrepancies and returns a structured, cited report. It is the same category of feature as ChatGPT’s Deep Research or Claude’s Research, but free.

Image generation

Qwen generates images in-app through Qwen-Image, at native 2K (2048×2048) resolution and with prompts up to roughly 1,000 tokens. Its differentiator is accurate text rendering inside images — signage, posters and UI mockups come out legible, where many image models garble text. This is a feature DeepSeek and Kimi do not offer at all.

Video and slide creation

The app turns a prompt into short videos (powered by Alibaba’s Wan models) and can assemble slide decks. Alibaba markets Qwen as “the best personal AI assistant,” and the video and slide tools are central to that positioning — most free chat apps stop at text and images.

Web Dev

Web Dev generates a complete, working webpage from a natural-language description — clean code plus a modern design you can preview and refine. It sits alongside artifacts-style output, so results are interactive rather than static.

Thinking mode

Thinking mode exposes Qwen’s full chain-of-thought: you can watch it plan, work through edge cases and reason step by step before answering. It is the equivalent of Claude’s extended thinking or DeepSeek’s DeepThink, and it markedly improves maths, logic and multi-step coding.

Web search

Qwen searches the web when a query needs current information and returns answers with source links, available free to all users.

Voice and multimodal

On mobile, Qwen supports voice and video chat for hands-free use, built on the Qwen3-Omni multimodal stack. The app understands text, images, audio and video together, so you can ask questions about an uploaded photo, document or clip in the same conversation.

Multi-platform

Qwen Studio runs on the web (chat.qwen.ai), iOS, Android, macOS and Windows, with conversation history synced across devices. The standalone consumer app launched in public beta in November 2025; before that, Qwen was used mainly through the web chat and Alibaba Cloud.

Strengths

Genuinely free, genuinely capable — Qwen pairs a near-frontier flagship with image generation, video, deep research and slide creation at no cost. Few free apps match its feature breadth; rivals gate most of this behind $20-a-month plans.

Multilingual reach — Qwen supports 119+ languages and was built with strong Chinese and broad multilingual coverage, making it one of the better free options for translation and non-English work.

Multimodal in one place — text, images (in and out), video, audio and documents are handled in a single app. Unlike DeepSeek, which cannot generate images, Qwen creates them.

Open-weight escape hatch — Qwen 3.5 and 3.6 are Apache 2.0, so the same capability can be self-hosted. The Qwen family is among the most-downloaded open-model lines in the world, with a large derivative-model ecosystem on Hugging Face.

Scale and momentum — Qwen was the fastest-growing AI app globally in November 2025, reaching 18.34 million monthly active users within two weeks of its public beta, and has continued to climb through 2026.

Limitations

China-based provider — Qwen is built by Alibaba, a China-headquartered company. The international app is operated by Alibaba Cloud (Singapore), but the corporate ties and the less-transparent consumer privacy terms make it a poor fit for sensitive, regulated or work data. See the privacy section.

English is second to Chinese — Qwen is strongest in Chinese and bilingual Chinese-English work. For English-first nuance and the hardest reasoning, US frontier models such as Claude Opus 4.8 and GPT-5.5 still lead.

No consumer paid tier for heavier guarantees — because the app is free with no paid plan, there are no priority-access or SLA options; for guaranteed throughput you move to the Alibaba Cloud API.

Vendor benchmarks run hot — Alibaba’s headline coding and agent scores are vendor-run and sit above independent leaderboards. Anchor expectations on the third-party Artificial Analysis index, not the launch slides.

No desktop autonomous agent — Qwen has deep research and tool use, but nothing equivalent to Claude Cowork that completes multi-step work on your own files and apps.

Who should use Qwen?

Great for:

• Cost-conscious users and students — a free, near-frontier assistant with image, video and research built in
• Multilingual and bilingual users — 119+ languages, with particular strength in Chinese
• Anyone who needs images or video for free — Qwen generates both; DeepSeek and Kimi do not
• Researchers — Deep Research returns structured, cited reports at no cost
• Developers — evaluate Qwen in the app, then self-host the Apache 2.0 open weights

Look elsewhere if:

• You handle sensitive or regulated data — use a Western-hosted app such as Le Chat, or self-host Qwen’s open weights
• You want best-in-class English reasoning — Claude or ChatGPT lead here
• You need enterprise compliance and data residency — go through Alibaba Cloud Model Studio’s regional endpoints or self-host
• You want a desktop agent that works on your files — Claude Cowork has no Qwen equivalent

Historical timeline

Privacy and data handling

Qwen’s privacy picture is more nuanced than “China app,” but it still warrants caution.

Who runs it. The international Qwen service is made available by Alibaba Cloud (Singapore) Private Limited, with terms governed by Singapore law and the courts of Singapore. That is a meaningful step removed from DeepSeek, whose app stores data directly on servers in mainland China. The mainland-China Tongyi version, however, stores data in China.

Training data. Alibaba Cloud states it will never use customer data to train models — but that promise is attached to the Model Studio API and enterprise platform, not necessarily the free consumer app. Qwen Studio’s consumer terms are less explicit about whether chats are used to improve models, so treat consumer-app inputs as potentially retained.

What the app collects. The Qwen Android listing’s Google Play “Data safety” section reports no third-party data sharing, collection of personal info and device identifiers, encryption in transit, and support for deletion requests.

The broader concern. Because Alibaba is China-based, analysts and several governments have raised concerns about the potential for state access to data held by Chinese AI providers — the same concern that has driven DeepSeek bans on official devices in some jurisdictions. For low-stakes personal use this is a manageable trade-off; for sensitive, regulated or work data it is not.

The safer path. Qwen 3.5 and 3.6 are Apache 2.0 open weights. Self-hosting them — via tools such as Ollama, llama.cpp or LM Studio, or through a non-Chinese cloud — gives you Qwen’s capability without sending anything to Alibaba. This is the recommended route for anyone with data-sovereignty requirements.

Qwen vs alternatives

Quick comparison

Where Qwen wins

• Free multimodality — image and video generation that DeepSeek and Kimi lack entirely
• Multilingual breadth — 119+ languages, strong on Chinese and bilingual work
• Feature density for $0 — deep research, slides and Web Dev in a free app
• Open-weight option — Apache 2.0 self-hosting for data control

Where competitors win

• DeepSeek — slightly cheaper, MIT-licensed open weights, and a coding-focused reputation, though no image generation
• ChatGPT and Claude — stronger English reasoning, deeper ecosystems, Western data hosting, and autonomous agents (Cowork)
• Le Chat — EU-based and GDPR-native, the better hosted choice for privacy-sensitive users

Getting started

Web

1. Go to chat.qwen.ai
2. Sign in
3. Pick a model and start chatting — no setup or payment required

Mobile

• iOS: download the Qwen (Qwen Studio) app from the App Store
• Android: download from Google Play or the regional app store

Desktop

Qwen Studio has dedicated macOS and Windows apps, downloadable from qwen.ai/download.

Tips for new users

Use the right mode for the job. Switch on Thinking mode for maths and code, Deep Research for multi-source questions, and the image or video tools for visual work — each routes to a model tuned for that task.

Lean on the languages. If you work across languages, Qwen is one of the strongest free options; ask it to translate or draft directly in the target language.

Keep sensitive data out. For confidential or regulated material, self-host the open weights rather than using the hosted app.

Frequently asked questions

Is Qwen free?

Yes. The Qwen app — now branded Qwen Studio, at chat.qwen.ai — is completely free with no subscription. It includes the flagship Qwen3.7-Max model, image and video generation, deep research and 119+ languages. You only pay if you build on Qwen through the Alibaba Cloud Model Studio API, which is billed per token.

What is Qwen Studio?

Qwen Studio is the new name for the Qwen Chat app; Alibaba renamed it on 10 April 2026. It is the same free assistant — web, mobile and desktop — giving everyday users and developers unified access to Qwen’s open-source and proprietary models, with chat, image and video generation, document understanding, voice chat and artifacts.

Who makes Qwen?

Qwen is made by the Qwen team at Alibaba Cloud, part of Alibaba Group, headquartered in Hangzhou, China. The app is one of the world’s most-used chatbots and runs Alibaba’s Qwen family of foundation models.

Can Qwen generate images and video?

Yes. Qwen generates images through Qwen-Image (native 2K resolution, with accurate in-image text) and short videos through Alibaba’s Wan models, both free in the app. This is a clear advantage over DeepSeek and Kimi, neither of which generates images.

Is Qwen safe to use for private or work data?

Use caution. Qwen is built by a China-headquartered company; the international app is run by Alibaba Cloud (Singapore), but consumer privacy terms are less transparent than Western rivals’, and the mainland version stores data in China. For sensitive, regulated or work data, prefer a Western-hosted app such as Le Chat, or self-host Qwen’s Apache 2.0 open weights so nothing leaves your infrastructure.

How good is Qwen compared to ChatGPT and Claude?

Qwen’s flagship Qwen3.7-Max is near-frontier — about 80.4% on SWE-bench Verified (Alibaba-reported) and a top-ten independent Artificial Analysis intelligence score — and free. On the hardest English reasoning and long-horizon agentic tasks, Claude Opus 4.8 and GPT-5.5 still lead, but the gap is much smaller than the price difference. See our best AI models ranking for the full picture.

Is Qwen open source?

Partly. Qwen’s mid-tier models — including Qwen 3.5 and Qwen 3.6 — are released under the Apache 2.0 licence and can be downloaded and self-hosted. The flagship Qwen3.7-Max is closed-weights and API-only, though it is free to use inside the app.

What languages does Qwen support?

Qwen supports 119+ languages and dialects, with particularly strong Chinese and broad multilingual coverage, making it one of the better free options for translation and non-English tasks.

Verdict

Qwen is the most generous free AI app in the market. It pairs a near-frontier flagship — Qwen3.7-Max, at roughly 80.4% on SWE-bench Verified and a top-ten independent intelligence score — with image generation, video, deep research, slide creation and Web Dev, all at no cost and across web, mobile and desktop. For multilingual and multimodal work on a budget, nothing else free comes close.

The hard limit is trust, not capability. Qwen is built by Alibaba, a China-headquartered company; the international app is Singapore-governed, but the consumer privacy terms are thin and the corporate ties are real. For low-stakes personal use, Qwen is an excellent free assistant. For sensitive, regulated or work data, use a Western-hosted alternative such as Le Chat, or take Qwen’s best feature — its Apache 2.0 open weights — and self-host, getting the capability without the data risk.

Official links