AL
Alibaba Group / Qwen team
Developer and publisher of Qwen3.8-27B and Qwen3.8-2.4T-A95B; released weights on Hugging Face and ModelScope under Apache 2.0 (27B) and a custom license (2.4T Max), positioning the release as evidence of Chinese open-weight AI competing with closed US frontier labs.
Published a Day 0 blog confirming Qwen3.8-27B runs on Ryzen AI Max and Radeon GPUs, positioning its hardware for local inference of the new model.
Published a technical blog on serving the larger Qwen3.8-2.4T-A95B model with configurable reasoning on GB300 NVL72 hardware, positioning its data-center GPUs as the inference platform for the flagship variant.
Third-party quantizer producing GGUF and NVFP4 versions of Qwen3.8-27B to enable local, lower-VRAM deployment; unofficial and not maintained by Alibaba.
Community critic who initially flagged what it read as a geographic license prohibition (USA, EU, UK, Korea); the claim was later found false, with the Apache 2.0 license containing no territorial restriction.
VA
Vals AI / infrastructure providers
Independent benchmark and infrastructure evaluators; Vals flagged that Alibaba modified benchmark timeouts versus Vals' own standardized settings, while several inference providers announced same-day serving support for the new models.