Alibaba's Qwen releases 27B open model, outperforms larger rival

Alibaba's Qwen team released open-weights models for Qwen3.8, marking a significant open-source AI release. The flagship model, Qwen3.8-27B, is a multimodal dense model with 27 billion parameters. According to Qwen, it outperforms the larger Qwen3.7-Plus in coding and office tasks—a notable achievement given its smaller size. The team also highlighted improved agent capabilities, with the model planning more independently and completing tasks more reliably.

The Qwen3.8-27B natively handles up to 262,000 tokens of context and can scale to one million tokens using the YaRN method, allowing it to process very long documents and multi-turn conversations. Beyond text, the model is multimodal and processes images, videos, diagrams, documents, and multi-hour video content. A flexible thinking mode is enabled by default but can be toggled on a per-query basis, giving users control over the model's reasoning depth.

Qwen also released weights for Qwen3.8-2.4T-A95B, a much larger model built for Max-level operation. Both models are distributed under the Apache 2.0 license, a permissive open-source license that permits free use and modification. The models are available immediately on Hugging Face and ModelScope. To serve users who prefer managed deployment, Qwen announced that a hosted version with one million tokens of context will soon be available through Qwen Cloud, Alibaba's AI service, offering an alternative for teams that want the model's capabilities without self-hosting infrastructure.