16GB of VRAM isn't really enough for coding or research. Take a look at GLM5.2, you would need at least 512GB of VRAM. You can run 20B models or Gemma4 for everyday questions, but these tasks don't really benefit from AI IMO.

Gemma4 could also run on a phone, which consumes much less power than an AMD BC-250, which uses about 120 to 350 W.