Qwen3.8-27B
Qwen3.8-27B is an open-weight, dense vision-language model released by Alibaba in mid-August 2026. It is a key member of the Qwen3.8 generation, designed to balance high-level reasoning capabilities with deployment-friendly efficiency.
Key Features
- Multimodal Capabilities: It is a native vision-language model, meaning it is trained to process text, images, and video (up to hour-long clips) natively, rather than relying on bolted-on adapters.
- Performance: Despite its 27-billion-parameter size, it is noted for matching or exceeding the performance of much larger models, such as the MoE-based Qwen3.7-Plus, in coding, research, and long-horizon agentic tasks.
- Flexible Thinking Control: The model includes a
reasoning_effortparameter that allows users to toggle the depth of its "thinking" process (xhigh,medium, orlow), making it adaptable for both complex analysis and fast, efficient responses. - Context Window: It features a native context window of 262,144 tokens, which can be extended up to 1,000,000 tokens using YaRN scaling.
- Accessibility: Released under the Apache 2.0 license, the model is designed to be runnable on consumer-grade hardware (such as high-end laptops or single GPUs) when quantized, making it highly practical for self-hosting.
Context in the Qwen Series
Qwen3.8-27B was released alongside the flagship Qwen3.8-2.4T-A95B (a massive 2.4-trillion-parameter sparse MoE model). While the flagship model represents the peak of Alibaba's current architectural capabilities, the 27B version is specifically optimized for developers and researchers who need high performance in a more manageable, deployment-friendly package.