Alibaba's Qwen team releases Qwen 3.8 models with open weights under the Apache 2.0 license

Details

Alibaba's Qwen team released open weights for two new models: Qwen3.8-27B, a 27-billion-parameter dense multimodal model, and the much larger Qwen3.8-2.4T-A95B for maximum capability workloads. Both ship under the Apache 2.0 license, making commercial use, fine-tuning, and redistribution unrestricted. Qwen3.8-27B handles text, images, and video — including diagrams, documents, and multi-hour video content. It natively processes 262,000 tokens and can scale to one million tokens using the YaRN extension method. Alibaba claims the model outperforms the larger Qwen3.7-Plus on coding and office tasks. A flexible thinking mode is on by default but can be toggled per individual query, giving developers control over reasoning depth versus latency. Both models are immediately available on Hugging Face and ModelScope. A hosted API with one million token context is coming to Qwen Cloud. The 27B parameter size hits a practical deployment sweet spot — competitive on complex tasks while fitting in multi-GPU consumer or mid-range cloud setups. The Apache 2.0 license is the key commercial differentiator: it enables enterprise fine-tuning without royalties or usage restrictions, putting real downward pressure on proprietary API pricing.