Skip to content
AITrendTool

Qwen

Alibaba's open-weight LLM family plus a free multimodal chat assistant and token-based API

Qwen is Alibaba's family of large language models — open-weight releases under the Apache 2.0 license, including the 235B-parameter Qwen3 mixture-of-experts model — paired with a free multimodal chat assistant and a token-based API. Qwen Chat is free to use; the API runs from about $0.05 per million input tokens via Alibaba Cloud. Best for developers who want capable open models to self-host or cheap, OpenAI-compatible API access.

Verified JUN 22, 2026 OPEN-SOURCE Live
Visit qwen.ai ↗ AI Chatbots
Screenshot of Qwen

What is Qwen?

Qwen is Alibaba’s family of large language models, and it shows up in three forms. First, there is Qwen Chat — a free multimodal assistant on the web and mobile that handles text, understands images and video, and generates images, with very large context windows on its top models. Second, the models are released as open weights under the permissive Apache 2.0 license, so anyone can download and self-host them at no license cost. The lineup spans from tiny models that run on modest hardware up to the Qwen3 mixture-of-experts flagship, which has 235 billion total parameters (with 22 billion active per token).

Third, for teams that would rather not run their own infrastructure, Qwen is available as a token-based API through Alibaba Cloud Model Studio, with OpenAI-compatible endpoints. Pricing is competitive: the lightweight Qwen-Flash starts around $0.05 per million input tokens, while the flagship Qwen3-Max starts near $1.20 per million, with discounts for batch processing and context caching. New accounts get a 90-day allowance of one million free tokens per model. The main caveats are that API pricing is tiered by context length and reasoning mode, and that prices and availability differ by region.

Who is it for?

Qwen appeals to people who value openness, low cost, and multilingual capability. The free chat is fine for everyday use, but Qwen’s real distinction is for builders who want to own their stack or keep API spend low.

  • Developers and startups who want a cheap, OpenAI-compatible API for high-volume workloads.
  • Self-hosting teams who need open weights they can run privately with no license cost.
  • Researchers and tinkerers who want a broad size ladder, from edge models to a 235B flagship.
  • Multilingual builders who need strong coverage across many languages in one model family.

How much does Qwen cost?

Starting price: $0 · Free tier: yes · Model: open-source

Pricing verified JUN 22, 2026

Price history tracked from June 2026

Qwen pricing tiers, verified against the official pricing page
Plan Price Includes
Qwen Chat Free Free multimodal web and mobile chat assistant · Text, image, and video understanding plus image generation · Context window up to roughly 1M tokens · No paywall or login wall for core use
Open weights Free Apache 2.0 licensed model weights · Qwen3 series, including the 235B-parameter mixture-of-experts model · Self-host via Hugging Face, Ollama, or vLLM · Full size ladder down to small edge models
API (Model Studio) $0.05/M Qwen-Flash from $0.05 per million input tokens · Qwen3-Max from $1.20 per million input tokens · OpenAI-compatible endpoints via Alibaba Cloud · 1M free tokens per model for 90 days on new accounts

What are Qwen's key features?

  • Free multimodal web and mobile chat (Qwen Chat)
  • Open-weight models under Apache 2.0, downloadable and self-hostable
  • Qwen3 mixture-of-experts flagship (235B total parameters) plus a full dense size ladder
  • Token-based API with OpenAI-compatible endpoints via Alibaba Cloud Model Studio
  • Context windows up to roughly 1M tokens on top models
  • Image generation and vision/video understanding
  • Thinking / reasoning modes for harder problems
  • Batch invocation (50% off) and context-caching discounts on the API

What people use Qwen for

  1. 01 Using a free, capable multimodal chat assistant for everyday questions and tasks
  2. 02 Self-hosting an open-weight model with no license cost for privacy or control
  3. 03 Calling a cheap, OpenAI-compatible API for high-volume text generation
  4. 04 Running a small Qwen model locally on edge or consumer hardware
  5. 05 Building multilingual applications across the model's broad language support

Pros and cons

Pros and cons of Qwen
Pros Cons
Genuinely free, capable, multimodal chat with no paywall for core use API pricing is tiered by context-length band and split into thinking/non-thinking rates, so cost is easy to underestimate
Truly open weights under Apache 2.0 — self-host with no license cost Ongoing API use is paid — the free quota is a 90-day, 1M-token-per-model trial for new accounts
Very cheap API at the low end (Flash from $0.05 per million input tokens) and competitive at the top Pricing and availability vary by deployment region (international vs mainland China)
Broad lineup from tiny edge models up to a 235B mixture-of-experts flagship Self-hosting the large mixture-of-experts models requires serious GPU resources

What are the best Qwen alternatives?

Frequently asked questions

Is Qwen free?

Yes, in two ways. The Qwen Chat assistant is free to use on the web and mobile, and the model weights are released under the Apache 2.0 license, so you can download and self-host them at no license cost. The paid part is the hosted API, billed per token.

What is the latest Qwen model?

The Qwen3 series is current, including the Qwen3-Max flagship and a 235B-parameter mixture-of-experts model (with 22B active parameters), alongside smaller dense models. The earlier Qwen2.5 generation is still widely used for self-hosting.

How much does the Qwen API cost?

It is token-based via Alibaba Cloud Model Studio. The cheapest model, Qwen-Flash, starts around $0.05 per million input tokens, while the flagship Qwen3-Max starts around $1.20 per million input tokens. New accounts get 1 million free tokens per model for 90 days.

Can I self-host Qwen?

Yes. The open-weight Qwen models are Apache 2.0 licensed and can be run via Hugging Face, Ollama, vLLM, and similar tooling. Smaller models run on modest hardware, while the large mixture-of-experts models need substantial GPU resources.

Is Qwen multimodal?

Yes. Qwen Chat handles text plus image and video understanding and can generate images, and several Qwen models support vision. Top models also offer large context windows (up to roughly 1 million tokens) and dedicated thinking modes for reasoning.

How does Qwen compare to DeepSeek and Llama?

Qwen, DeepSeek, and Llama are all open-weight model families. Qwen stands out for its broad size ladder, strong multilingual support, and a free hosted chat assistant. As with any model, the right choice depends on your task, language needs, and hosting constraints.

Public signals