Qwen Review (2026): Features, Pricing & Verdict
Alibaba’s Qwen is the most prolific open-weight AI family on Earth — dozens of free, downloadable models and a capable free chat app. Here’s an honest look at what makes it a developer favourite, and the privacy and censorship trade-offs that come with it.
While Western labs guard their best models behind APIs, Alibaba has spent three years doing the opposite — pouring frontier-grade models into the open under a permissive licence. Qwen (or Tongyi Qianwen) is the result: the largest, most complete open-weight model family in existence, paired with a free chat app at chat.qwen.ai. It’s a staple of the open-source community, genuinely strong at coding and reasoning, and astonishing on languages. It also carries the same privacy and censorship questions as any China-hosted AI. This review covers the consumer Qwen Chat app and Alibaba’s open Qwen models; the Alibaba Qwen API and model platform are reviewed separately.
The models in 2026
Qwen’s defining feature is breadth. The Qwen3 family alone spans eight models from 0.6B to 235B parameters, all released under the open Apache 2.0 licence, running on everything from a phone to an H100 cluster. The flagship Qwen3-235B sits neck-and-neck with the best open-weight generalists, Qwen3-Coder (480B) matches Claude Sonnet on agentic coding, and Qwen3.5 pushed the family to 397B parameters with support for over 200 languages.
The newest release, Qwen3.7-Max (May 2026), is pitched as an “agent frontier” model built for long-horizon, tool-heavy tasks with a one-million-token context. Notably, it’s API-only — not open-weight, a shift from Alibaba’s open tradition, while the open line continues at Qwen3.6 and below. Qwen Chat lets you use the latest preview models free, with a thinking mode for harder problems.
Features
Chat, thinking & search
Qwen Chat covers the essentials well: conversational chat, a reasoning “thinking” mode, web search for current information, and document processing for files you upload. Answers are strong on technical, mathematical and coding questions.
Vision, images & artifacts
The app handles image and video understanding, generates images, and supports artifacts — interactive outputs you can preview and refine — plus tool use. It’s a more multimodal experience than some open-source rivals offer out of the box.
Multilingual & open weights
Qwen’s standout is languages: the latest models support over 200, making it one of the best choices on the planet for non-English work. And because most Qwen models are open-weight, you can download them from Hugging Face and run them privately on your own hardware.
Pricing in 2026
Qwen is among the cheapest routes to a capable model anywhere, and the chat app is free.
| Plan | Price | What you get |
|---|---|---|
| Qwen Chat | $0 | Free account access to the latest Qwen models with thinking mode, web search, vision, image generation and artifacts, subject to rate limits |
| Paid consumer tier | None | There is no consumer subscription — the app is free |
| API (DashScope) | From ~$0.01/1M | Pay-per-token via Alibaba Cloud Model Studio; ~$0.78/1M for Qwen3 Max, with free credits for new accounts (reviewed separately) |
| Self-hosted | Free (your hardware) | Apache 2.0 open weights you can download and run locally for full data control |
For developers, the combination of free open weights and a near-free API is hard to beat. For everyday users, the free chat app is generous — the main limits are rate caps and, as below, the privacy and censorship trade-offs of a China-hosted service.
Pros and cons
✅ Strengths
- The largest, most complete open-weight family (0.6B–397B)
- Apache 2.0 licence — free to download and self-host
- Excellent coding (Qwen3-Coder rivals Claude Sonnet)
- Best-in-class multilingual support (200+ languages)
- Free chat app with vision, image gen and artifacts
- Extremely cheap API; runs from phones to clusters
- Backed by Alibaba, a top-tier global AI lab
❌ Weaknesses
- Hosted app stores data in China under Chinese law
- Censors politically sensitive topics
- Newest flagship (3.7-Max) is closed, not open-weight
- Consumer app less polished and less known in the West
- English writing trails the top Western models
- Rapid versioning (3.5, 3.6, 3.7) is confusing
Where it falls short
As with any China-hosted assistant, the first concern is data privacy. The Qwen Chat app is run by Alibaba, so your conversations are stored in China and subject to Chinese law. For general, non-sensitive work that’s fine; for confidential, client or personal data, it isn’t the right tool. The saving grace is the open weights — self-host a Qwen model and the data never leaves your hardware, which is exactly why so many privacy-conscious developers use Qwen this way.
The second is censorship: in line with Chinese content rules, Qwen declines or deflects on politically sensitive subjects, which matters if you value unfiltered inquiry. Beyond those, there are smaller caveats. Alibaba’s newest flagship has gone closed, a step back from its open ethos; the consumer Qwen Chat app is less refined and far less known outside China and the developer world; its English prose is good but not quite at the level of the best Western models; and the rapid-fire versioning (3.5 to 3.6 to 3.7 in months) makes it hard to track what’s current. None of this dims Qwen’s value as an open-weight powerhouse — it just defines who should use it, and how.
Scorecard
How Qwen scores across what matters, averaging to 7.4/10:
Verdict
Qwen is the most generous open-source story in AI, and for the right person it’s outstanding. If you build on open models, self-host for privacy, work across many languages, or want frontier-adjacent coding for almost nothing, Qwen is one of the best options in the world — and the free chat app is a capable bonus. Alibaba’s open-weight strategy has genuinely shaped the whole field.
It scores 7.4 because the trade-offs are real: the hosted app’s data goes to China, sensitive topics are censored, the newest flagship has closed up, and the consumer experience trails the polished Western apps. Use it freely for non-confidential technical and multilingual work — ideally self-hosted — and keep confidential data on a tool you trust. Its closest sibling is covered in our DeepSeek review; for another open-weight option with stronger privacy see our Le Chat by Mistral review, and for the all-round leaders our ChatGPT review and Claude review.
Frequently asked questions
Is Qwen free?
Yes. The Qwen Chat app at chat.qwen.ai is free with a registered account, including the latest models, thinking mode, web search, vision, image generation and artifacts, subject to rate limits. The API is cheap but billed by usage.
Is Qwen open source?
Mostly. Most Qwen models — including the Qwen3 family up to 235B and Qwen3.6 — are released under the open Apache 2.0 licence and can be downloaded and self-hosted. The newest flagship, Qwen3.7-Max, is an exception: it’s API-only.
What models does Qwen use?
The Qwen3 family (0.6B to 235B), Qwen3.5 (397B, 200+ languages), the Qwen3-Coder coding model, and the newest Qwen3.7-Max agent model, plus vision and image-generation models — most open-weight, the latest flagship closed.
Is Qwen private and safe to use?
The hosted app stores data in China under Chinese law, so treat it as fine for non-sensitive work but unsuitable for confidential data. Self-hosting the open-weight models avoids the data-residency issue entirely.
Does Qwen censor topics?
Yes. In line with Chinese content rules it declines or deflects on politically sensitive subjects. This rarely affects technical or everyday use but is a genuine limitation for open inquiry.
Is Qwen better than ChatGPT or DeepSeek?
For open-weight breadth and multilingual work it’s arguably the best, and on coding it’s excellent. ChatGPT is more polished and broader; DeepSeek is a close open-weight sibling. Like DeepSeek, Qwen carries China-hosting privacy and censorship caveats.
Reviewed June 2026 by AINewsAndUpdates.com. Models, features and pricing change quickly — always confirm current details on the official site before relying on it.