Alibaba (Qwen) Open Models: Model List, Licenses and Articles

September 26, 2026September 27, 2026

About Alibaba (Qwen)

Qwen is the family of large language models developed by Alibaba Cloud. The Qwen organization on Hugging Face describes itself as “the organization of Qwen, which refers to the large language model family built by Alibaba Cloud", and it keeps releasing language models (LLMs), multimodal models (LMMs) and related projects. The license of the image model Qwen-Image-2.1 names Hangzhou Tongyi Laboratory Technology Co., Ltd. (Tongyi Lab) as the licensor.

In the local LLM world, Qwen is one of the publishers whose models are most often used as a base for other models. In our own coverage, articles on third-party fine-tunes and quantized builds of Qwen models outnumber the articles on Qwen’s own releases (see the tables below).

What It Releases

The list of recent models (release date, task, license, parameters) is in the “Recent Models on Hugging Face" table below, collected by code every day. Here are the main points.

• The size range is unusually wide, from 0.8B to 2.4T. Qwen3.5 alone came in eight sizes, covering everything from laptop-sized models to data-center MoE models in one generation.
• Qwen3.8 is the first time Qwen has released a “Max-class" model. The model card describes Qwen3.8-2.4T-A95B as bringing “a Qwen-Max-class model to open release" for the first time.
• Qwen3.8-Flash-Next is an experimental preview of the architecture behind the next generation (Qwen4), according to its model card. It tries out a pairing of linear attention with a new sparse attention (QSA), n-gram embeddings and more.
• Video generation research (ali-vilab): this page also covers the ali-vilab organization, whose organization card states “Alibaba TongYi Vision Intelligence Lab". It publishes VACE (April 2025, for Wan2.1 and LTX-Video), which unifies video generation and editing, and early text-to-video models.
• For benchmarks and architecture details of each family, follow the family pages from the “By Model Family" table below.

Licenses Differ from Model to Model

Not every Qwen model is Apache-2.0. Its 2026 models use at least four different licenses.

• Apache-2.0 (Qwen3.5, Qwen3.6, Qwen3.8-27B and others): broad use, including commercial use.
• Qwen3.8-Max License (Qwen3.8-2.4T-A95B): free, including commercial use, with two conditions. Products with more than 100 million monthly active users or more than US$20 million in monthly revenue must display the model name prominently. A licensee running a Model-as-a-Service business (offering the model through an API and the like) or an AI Work Assistant business (AI products for coding or office productivity) with more than US$50 million in revenue over twelve consecutive months must obtain a separate license before commercial use (internal use that is not offered to third parties is exempt).
• Qwen Community License 1.0 (Qwen3.8-Flash-Next): nearly the same text as the Qwen3.8-Max License, but commercial use in a Model-as-a-Service or AI Work Assistant business requires a separate license regardless of revenue.
• Qwen Research License (Qwen-Image-2.1): permitted for research or evaluation (non-commercial) purposes only. Commercial use requires a separate license.

Running any of these models on your own machine to try them out is allowed under all four licenses. Commercial uses such as selling generated images or offering the model as an API are treated very differently depending on the license. This summary is not legal advice; read each model’s license text before use.

Running Them Locally

• Official quantized builds are mainly FP8 and GPTQ-Int4. Qwen itself publishes FP8 and GPTQ-Int4 versions of Qwen3.5, Qwen3.6 and Qwen3.8-27B, formats meant for GPU servers such as vLLM and SGLang.
• For recent families, GGUF builds come from third parties. The last official GGUF we can find on the Qwen organization is Qwen3-Coder-Next-GGUF from February 2026; there are none for Qwen3.5 and later (as of 2026-09-26). For llama.cpp, Ollama and LM Studio you will use conversions by quantizers such as unsloth and bartowski.
• Derived models are not the official models. Fine-tunes, uncensored versions and ternary rebuilds based on Qwen models can differ from the original in both weights and license. The “Fine-tunes and Builds by Others" table below lists them separately from the articles on Qwen’s own models.

Sources: the organization cards of Qwen and ali-vilab; model cards and license texts of Qwen/Qwen3.8-27B, Qwen/Qwen3.8-2.4T-A95B, Qwen/Qwen3.8-Flash-Next and Qwen/Qwen-Image-2.1; the model list of the Qwen organization on Hugging Face (all as of 2026-09-26).

Our Coverage and Data

Local Model Watch has published 2 article(s) on Alibaba (Qwen)’s own models, 12 on third-party fine-tunes and quantized builds based on them, and 0 from its official blog. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.

At a Glance

Recent Models on Hugging Face

The publisher lists 480 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.

Qwen (468 models)

ali-vilab (12 models)

From the Hugging Face API. Last changed 2026-09-27 (JST). Collected by code.

Licenses of Its Own Models (as recorded in our articles)

The license can differ from model to model even within one publisher. This is not legal advice; check each model’s license text before use.

By Model Family

Articles on Its Own Models

Fine-tunes and Builds by Others

Last updated 2026-09-27 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.