Qwen3.8 27B

Qwen3.8-27B is a dense 27.78-billion-parameter multimodal foundation model for coding, professional work, research, and long-horizon agents. It natively understands text, images, and video, supports configurable thinking effort and preserved reasoning across turns, and uses a 64-layer hybrid architecture with repeating Gated DeltaNet and full-attention blocks. Its native context window is 262,144 tokens and can be extended to approximately 1 million tokens. The open weights are released under Apache 2.0 and support Transformers, vLLM, SGLang, and TokenSpeed.