Qwen3.8-27B open-weight model launches, tops Hugging Face

Alibaba's Qwen3.8-27B, a 27B-parameter multimodal dense model with 262K native context, outperforms Qwen3.7-Plus and scores 52 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Luna. It runs locally on consumer hardware, with Simon Willison noting it defaults to 'xhigh' reasoning effort causing overthinking.