Qwen3.8-27B lands as Alibaba's compact open vision-language model

The Qwen team has released Qwen3.8-27B on Hugging Face under Apache-2.0, and it has picked up around 17,800 likes in about two months. It's a dense 27B model with a native vision encoder, 262K-token context natively (extensible to 1M), built on the Gated DeltaNet hybrid architecture from Qwen3.5 with multi-token prediction. The model card's numbers are strong for the size: 61.7 on SWE-bench Pro, 73.0 on Terminal-Bench 2.1, 90.3 on LiveCodeBench v6, and 84.3 on OSWorld-Verified, alongside 94.6 on MathVision with a code interpreter. Thinking mode is on by default with adjustable reasoning_effort, and a hosted version on Qwen Cloud with 1M context and built-in tools is announced but not yet live. Community builds are already spreading — NVIDIA's NVFP4 variant and several Apple-Silicon MLX quantizations are up.

Summarized by the Neuricon wire. Every claim links back to its primary source where available.