On the evening of August 14, Alibaba's Qwen large model team officially announced the open-sourcing of the Qwen3.8-27B model, a significant addition to the Qwen3.8 family. Distributed under the Apache 2.0 license, this model grants all developers, research institutions, and enterprises unrestricted access to its weights, enabling them to freely download, deploy, and utilize it for commercial purposes. Qwen3.8-27B is a native multimodal dense model boasting 27 billion parameters and inherently supports a context length of 262K tokens. This capacity can be further expanded to 1M tokens using YaRN technology. Additionally, the model incorporates a novel reasoning_effort feature, which intelligently adjusts the depth of reasoning based on the complexity of the task, thereby optimizing computational resource usage. In terms of performance, Qwen3.8-27B demonstrates marked improvements over its predecessor, Qwen3.6-27B, particularly in programming and office applications, achieving an overall performance level that surpasses Qwen3.7-Plus. The quantized version of the model operates seamlessly on consumer-grade graphics cards. Presently, Qwen3.8-27B is accessible on both Hugging Face and ModelScope platforms.