Alibaba Unveils Qwen3.8-27B and Releases Weights of Qwen3.8 Flagship Model

Within just two days of release, Qwen3.8-27B has become one of the top 5 most-liked models on Hugging Face.
Alibaba has open-sourced more than 460 models, spawning over 300,000 derivative models and accumulating more than 3 billion global downloads

Alibaba has unveiled Qwen3.8-27B, the latest open-weight addition to its Qwen3.8 series. This native multimodal dense model balances high performance with cost-efficiency. It can run on consumer-grade hardware, and with quantization, it can even run on a laptop. Despite its compact footprint, Qwen3.8-27B delivers excellent capabilities in handling coding, professional work, research, and long-horizon agentic tasks, matching the performance of Qwen3.7-plus — an MoE model ten times its size.

As a native vision-language model, Qwen3.8-27B understands images and videos while maintaining flexible thinking control. Released under the Apache 2.0 license, it features a native 262K-token context window that easily extends to 1 million tokens. Furthermore, the model is designed to execute complex, multi-step tasks to completion with high reliability. Within just two days of release, the model has become one of the top 5 most-liked models on Hugging Face.

Alibaba also released the weights of its flagship Qwen3.8 large language model, Qwen3.8-2.4T-A95B, on Hugging Face and ModelScope (an Alibaba-initiated open-source community), making it freely available for download by global developers and businesses. It is also the first time that Qwen3.8 brings a Qwen-Max-class model to open release. As the most powerful Qwen model to date, this flagship version features 2.4 trillion total parameters with 95 billion active parameters, and supports a context length of up to 1 million tokens.

On independent LLM evaluation platforms, the flagship model ranked third globally on Arena AI's CodeArena (which evaluates models on front-end web development tasks), and third on Artificial Analysis' Agentic Index (which evaluates agentic capabilities).

Across both Hugging Face and Model Studio, Alibaba has open-sourced more than 460 models. This ecosystem has spawned over 300,000 derivative models and accumulated more than 3 billion global downloads, making it the world's most-downloaded open-source model family.

According to Hugging Face data shared in the blog "State of Open Models: Summer 2026 Observations", Qwen has become the community's base model. Qwen-based models now account for 151,448 derivatives on the Hub (which is 2.6 times Meta's total footprint and 4.7 times that of Llama repositories specifically), earning 2,045 million downloads (across repositories with declared parameter counts) on the Hub from January to July this year.

Thanks to its consistency (maintaining a regular release cadence), coverage (publishing models across a wide range of sizes and use cases), and openness (an Apache 2.0 license that reduces friction for modification, redistribution, and commercial use), “Qwen has become part of the default workflow for developers deciding what models to fine-tune and deploy,” the blog noted.

Data from Hugging Face

The company has been contributing to the open-source community for a long period of time.Recently, to make any agent framework natively multimodal,Alibaba open sourced Qwen-MM-Plugins, a harness extension library designed for multimodal agents, providing agent frameworks with image and video processing, multimodal memory, dynamic-resolution support, visual tool use, and specialized capabilities for tasks such as video editing.

This article was originally published on Alizila written by Crystal Liu.

Community

Alibaba Unveils Qwen3.8-27B and Releases Weights of Qwen3.8 Flagship Model

Read previous post:

Alibaba Cloud Community

You may also like

Comments

Alibaba Cloud Community

Related Products

Token Plan

Alibaba Cloud Model Studio

Qwen

Managed Service for Prometheus