[Qwen3.8-27B Open Source: Runs Locally with 17GB After Quantization] According to monitoring by Beating, Qwen has officially released the Qwen3.8-27B weights. This is the version of the Qwen3.8 series truly designed for local deployment, utilizing a 27B dense architecture. The official model card indicates that it natively supports image and video understanding, with a context length of up to 262,144 tokens, expandable to approximately 1 million tokens. The hardware requirements are particularly noteworthy. Unsloth has already provided quantization support, stating that Qwen3.8-27B can run locally on approximately 17GB of RAM or VRAM after quantization. This means that a 24GB consumer-grade GPU, or even some high-memory Macs, can run Qwen3.8 on their own machines. Previously released Qwen3.8 large model weights with 2.4T parameters only supported text and required enabling "thinking mode." In contrast, the 27B version retains support for image and video input. For most individual developers, the 27B version is the truly practical and directly usable open-source release in this round of Qwen3.8 updates. [Original Link]