Qwen (short for Tongyi Qianwen) is the family of AI models made by Alibaba. It's best known as the world's most popular open-weight AI: most Qwen models are released under the permissive Apache 2.0 license, they're the most-downloaded large language models on Hugging Face, and they cover a huge range of sizes and languages. You can use Qwen through Alibaba's app and API, or download the open models and run them yourself for free.

Qwen is developed by Alibaba, the Chinese technology and e-commerce giant, through its Tongyi (Qwen) team and Alibaba Cloud. It's Alibaba's answer to models like GPT and Claude, and one of the flagship efforts of China's AI industry — particularly influential because Alibaba open-sources most of the family.

Largely, yes. Most Qwen models are open-weight under Apache 2.0, which means you can download and run them on your own hardware at no per-token cost — the real cost floor is zero. Alibaba also offers a free consumer chat app and a paid hosted API (via Alibaba Cloud) that's inexpensive, ranging from a few cents per million tokens on the small open models up to a few dollars on the closed Qwen-Max flagship.

Most of it, yes. The bulk of the Qwen3 family — from tiny 0.6B models up to the large 235B mixture-of-experts model — is released as open weights under the Apache 2.0 license, which is why Qwen has become the de facto standard for open-weight AI. Alibaba keeps its very top frontier model, Qwen-Max, closed and API-only, but the open models cover the vast majority of real-world use.

For everyday use it's fine, and Qwen has a real safety advantage: because most of it is open-weight, you can self-host it and keep your data entirely on your own infrastructure — no data leaves your machine. If you instead use Alibaba's hosted API or app, you're subject to Alibaba Cloud's terms and Chinese data jurisdiction, and, like all Chinese models, Qwen applies content rules on politically sensitive topics. For confidential work, self-hosting the open weights is the safe route.

What Is Qwen? Alibaba's Open-Source AI Explained (2026)

Q: Qwen vs Doubao — what's the difference?

They're both from Chinese giants but play different games. Qwen (Alibaba) is the global open-source standard — open weights, multilingual, used by developers everywhere. Doubao (ByteDance) is China's most-used consumer AI app, focused on the domestic mass market and largely closed. If you're a developer or want to self-host, Qwen; if you want to know what hundreds of millions of Chinese consumers actually use day to day, that's Doubao.

What is Qwen — Alibaba's open-source AI explained

Qwen (short for Tongyi Qianwen) is Alibaba's family of AI models — and the world's most popular open-weight AI. Most of the family is released for free under the permissive Apache 2.0 license, the models are the most-downloaded large language models on Hugging Face, and they span everything from tiny 0.6-billion-parameter models to a giant 235-billion mixture-of-experts. If DeepSeek is China's breakout model and Doubao is its consumer favorite, Qwen is its open-source gift to the world's developers. Here's what it is and why it matters.

Key takeaways

What it is: Alibaba's AI model family (Tongyi Qianwen), the de facto standard for open-weight AI.
Open-source: most Qwen models are Apache-2.0 and the most-downloaded LLMs on Hugging Face — free to self-host.
Range: from 0.6B up to a 235B mixture-of-experts, plus a closed frontier flagship, Qwen-Max.
Multilingual-first: designed for 100+ languages, one of its biggest strengths.
Cost: free to self-host; the hosted API is cheap, from a few cents to a few dollars per 1M tokens.

Who Makes Qwen?

Qwen is built by Alibaba — the Chinese technology and e-commerce giant — through its Tongyi (Qwen) team and Alibaba Cloud. "Tongyi Qianwen" (通义千问) is the full Chinese name; "Qwen" is the shorthand you'll see on model cards and APIs. It's Alibaba's answer to OpenAI's GPT and Anthropic's Claude, and one of the most influential efforts in China's AI industry — precisely because Alibaba chose to open-source most of it rather than keep it locked behind an API.

Why Qwen Matters: The Open-Weight Standard

Here's the single most important thing about Qwen: it's open. The bulk of the Qwen3 generation is released under the Apache 2.0 license — one of the most permissive open-source licenses there is — which means anyone can download the weights, run them, fine-tune them, and even use them commercially, for free. That openness has made Qwen the most-downloaded large-language-model family on Hugging Face and the default base model for countless other projects and startups worldwide.

The family is unusually broad. It ranges from tiny 0.6B models that run on a laptop, through mid-size dense models, up to a large 235B-parameter mixture-of-experts that competes with frontier models. Alibaba keeps only its very top model — the closed Qwen-Max, an API-only frontier tier with a 1M-token context window — behind a paywall. Everything else, you can just download.

Most Chinese labs compete on price or product. Alibaba competes on ubiquity: by giving Qwen away as open weights, it has made its models the foundation other people build on — which is a different, and arguably deeper, kind of influence.

What Can Qwen Do?

Qwen is a full general-purpose model family — chat, reasoning, coding, long-document analysis, and vision in its multimodal variants. Two strengths stand out:

Multilingual by design. Qwen was built "multilingual-first," trained across 100+ languages — including Korean, Japanese, Southeast Asian languages and Arabic — which makes it one of the strongest models for non-English work.
Coding and reasoning. The larger Qwen models are competitive with Western flagships on coding and math benchmarks, and there are coding-specialized variants in the family.

You can use Qwen three ways: through Alibaba's free consumer chat app, through the hosted API on Alibaba Cloud (Model Studio / DashScope), or by self-hosting the open weights — the same route we cover for other open-weight Chinese models like GLM.

Is Qwen Free?

Largely, yes. Because most Qwen models are open-weight, self-hosting them costs zero per token — you only pay for the hardware to run them. That's the real cost floor, and it's why Qwen is so heavily used. If you'd rather not run your own servers, Alibaba's hosted API is cheap: a few cents per million tokens on the small open models, up to a few dollars on the closed Qwen-Max flagship. There's also a free consumer chat app for casual use. (To price a specific workload, our Token Calculator covers the major models.)

Qwen vs Doubao: China's Two Giants, Different Games

Alibaba's Qwen and ByteDance's Doubao are both from Chinese tech giants, but they're aimed at completely different targets.

Qwen is the developer's model. Open-weight, multilingual, global — it's what people build on top of, all over the world.
Doubao is the consumer's app. ByteDance's Doubao is China's most-used AI assistant (260M+ monthly users), but it's largely closed and focused on the domestic mass market.

Put simply: Qwen is China's open-source export; Doubao is its domestic consumer champion. If you're a developer or want to self-host, Qwen is far more relevant. If you want to understand what hundreds of millions of Chinese users actually chat with, that's Doubao — which we break down in what is Doubao?

Is Qwen Safe?

Qwen has a genuine safety advantage over most Chinese models: because it's open-weight, you can self-host it and keep your data entirely on your own infrastructure. Nothing leaves your machine, no Chinese company sees your prompts, and most content filtering doesn't apply. That makes Qwen one of the more privacy-friendly options in the Chinese ecosystem — if you self-host.

If you instead use Alibaba's hosted app or API, the usual caveats apply: you're subject to Alibaba Cloud's terms and Chinese data jurisdiction, and Qwen applies content rules on politically sensitive topics like any Chinese model. We cover the full picture in are Chinese AI models safe?

Should You Use Qwen?

Use Qwen if you're a developer who wants a capable, free, open-weight model to build on or self-host, you need strong multilingual performance, or you want frontier-ish quality without a frontier price. Look elsewhere if you want a polished consumer product with a big ecosystem (Claude, ChatGPT) or the absolute top of reasoning capability. For a sense of where Qwen and the other Chinese models sit on price, see the 2026 AI API price war, and if you just want a recommendation for your use case, our free picker takes about 30 seconds.

Frequently Asked Questions

What is Qwen?

Qwen (Tongyi Qianwen) is Alibaba's family of AI models, best known as the world's most popular open-weight AI: most models are Apache-2.0 licensed, they're the most-downloaded LLMs on Hugging Face, and they span tiny to very large sizes. You can use it via Alibaba's app and API or self-host the open models for free.

Who makes Qwen?

Alibaba, through its Tongyi (Qwen) team and Alibaba Cloud. It's Alibaba's flagship AI and one of China's most influential model families, largely because it's open-sourced.

Is Qwen free?

Mostly. Open-weight Qwen models are free to self-host (zero per-token cost); Alibaba's hosted API is cheap (a few cents to a few dollars per 1M tokens), and there's a free consumer chat app.

Is Qwen open source?

Most of it — the Qwen3 family from 0.6B to 235B is Apache-2.0 open-weight, which made it the de facto open-weight standard. Only the top Qwen-Max flagship is closed and API-only.

Qwen vs Doubao — what's the difference?

Qwen (Alibaba) is the global open-source standard for developers; Doubao (ByteDance) is China's most-used consumer AI app, largely closed and domestic-focused. Developers and self-hosters want Qwen; Doubao is the consumer story.

Is Qwen safe to use?

Fine for everyday use, and safer than most Chinese models if you self-host the open weights — your data stays on your own hardware. Using Alibaba's hosted API means Chinese data jurisdiction and content rules apply, so self-host for confidential work.