Users with 80-160GB unified memory or high-bandwidth RAM face limitations due to the lack of models sized for their hardware. Existing models are either too small for performance or too large for memory constraints, prompting a call for 100B-scale sparse models like Qwen 3.5 122B or Gemma 4 122B to better serve users with AMD AI Pro, RTX 3090/5090, or Apple devices.
We need a 80-160B model urgently for unified memory devices
from English