Community discusses parameter-count ceiling for small model intelligence

A Reddit user praises the Qwen3.6-27b model but questions if smaller models face a hard intelligence ceiling due to parameter count or VRAM limits. Commenters debate whether improvements can continue or if diminishing returns set in at smaller scales.