llama.cpp · GPU · local inventory

Model Rack / rated & trimmed

Scored for smarts, footprint, decode speed, and tool-calling. Score = Artificial Analysis Intelligence Index (higher = smarter); dashed = my estimate. t/s = est. decode on a Ryzen AI MAX+ 395 (Q4). Tick boxes to compare models side by side.

36 in library 16 removed ~333 GB freed
≥45 35–44 25–34 15–24 <15 estimate
Show
Filter by capability
Sort
0 selected

Compare

Highlighted = best among selected (highest score, smallest disk, fastest decode).