Alibaba Open-Sources Qwen3.8-27B Model

The official WeChat account for Alibaba's "Qwen Large Model" announced today that the Qwen3.8-27B model has been officially open-sourced, allowing developers, research institutions, and enterprises to freely download, deploy, and use it. This 27-billion-parameter dense multimodal model supports a native context length of 262K tokens and can be extended up to 1 million tokens using YaRN technology. Compared to its predecessor, Qwen3.6-27B, Qwen3.8-27B demonstrates improved performance in programming and office-related scenarios and outperforms Qwen3.7-Plus. The new model introduces a reasoning_effort feature that dynamically adjusts reasoning depth based on task complexity to conserve computational resources. Inheriting the end-to-end capabilities of the Qwen3.8 series, Qwen3.8-27B is capable of handling complex tasks. The official team also released benchmark results showcasing the model’s performance across various text and vision scenarios.