On August 14, 2026, Alibaba Cloud's Qwen team declared the official open-sourcing of the Qwen3.8 series models. This move grants developers, research institutions, and enterprises across the globe the freedom to download, deploy, and utilize these models for commercial endeavors.

Among these models, Qwen3.8-27B stands out as a native multimodal dense model. It is equipped with 27 billion parameters and inherently supports a 262K context. Moreover, its technology facilitates extrapolation to 1M Tokens. When compared to Qwen3.7-Plus, its overall performance is superior, especially excelling in programming and office-related scenarios.

This model introduces a novel reasoning_effort feature. This feature can dynamically adjust the depth of thinking based on the difficulty level of the task, thereby conserving computational resources. After quantization, Qwen3.8-27B can operate seamlessly on consumer-grade graphics cards, substantially reducing the hurdles for local deployment.

In addition, the flagship model Qwen3.8-Max has also been open-sourced at the same time. Up to now, the Qwen series models have cumulatively open-sourced over 460 models. The total number of global downloads has surpassed 3 billion, and more than 300,000 derivative models have been created.