Alibaba releases Qwen3.8-27B open-weight model

Qwen3.8-27B is a native multimodal dense model with 27B parameters, 262K native context (extendable to 1M), and outperforms Qwen3.7-Plus overall. Inference costs $0.16 per million tokens, and it's Apache 2 licensed. It became the #1 trending model on Hugging Face.