Alibaba Ships 27B Model

HangzhouChina’s Alibaba released a 27-billion-parameter open model.

FP8 quantization halves memory, letting the model fit on one commodity data-centre GPU.

Qwen3.8-27B natively handles a 262,144-token context, extensible to 1,000,000, and reads images and video, its team said.

A hosted Qwen Cloud version with 1 million tokens by default is coming soon, the company said.

Sources: Hacker News