AMD adds day-one support for Alibaba’s open-source Qwen3.8-27B model

AMD said it is offering Day 0 support for Alibaba’s latest Qwen model, Qwen3.8-27B, allowing developers to run the large open-source AI model locally on AMD hardware on the day of release.

27B-parameter model positioned for local AI development

AMD described Qwen3.8-27B as a 27-billion-parameter dense model suited for local AI development. The company said it continues the Qwen series’ optimization path in code generation, practical work tasks, scientific research and long-context AI applications.

According to AMD, the model can run through the open-source inference framework llama.cpp on AI PCs and workstations equipped with AMD processors, or on a single AMD 32GB graphics card. It also supports AMD hardware platforms with more than 24GB of variable graphics memory, or graphics memory capacity.

Preliminary performance figures on AMD hardware

AMD said early testing showed relatively strong local inference performance for Qwen3.8-27B on its platform.

• On the AMD Ryzen AI Max+ 395 processor, performance reached as high as 24.5 tokens per second.
• On a single Radeon AI PRO R9700 GPU, performance reached as high as 51.8 tokens per second.

The tests were based on Windows, the llama.cpp Vulkan backend and multi-token prediction, or MTP, optimization. AMD said actual performance still has room to improve as later software and model optimization moves ahead.

Alibaba open-sources the Qwen3.8 series

Alibaba said today that it has officially open-sourced the Qwen3.8 model family, making it available for developers, research institutions and enterprises to download, deploy and use freely.

The newly open-sourced Qwen3.8-27B is a native multimodal dense model with 27 billion parameters. Alibaba said the model’s overall capability exceeds Qwen3.7-Plus.