Alibaba’s Qwen team posts a massive Qwen3.8 2.4T-parameter FP8 model checkpoint on Hugging Face

A Hugging Face listing points to Qwen3.8-2.4T-A95B-FP8, a very large Qwen model checkpoint distributed in FP8 format. The name suggests a 2.4T-parameter model with about 95B active parameters, though the linked page provides the primary source details.

The release signals continued pressure among leading AI labs to make frontier-scale or near-frontier open-weight models more accessible. FP8 distribution also reflects a broader trend toward lower-precision formats that reduce storage and inference costs for extremely large models.