Qwen3 4B vs 0.6B on a 6 GB GPU: Speed, Memory, and Answer Quality

For my personal retrieval-augmented generation (RAG) study, I needed a local model to answer...