Local small modelsrunning in your browser

Local Copy Enricher

An offline benchmark of 3 small models on 15 products. Explore the comparison, the actual generated copy, the cost study, and determinism — all from committed results.

modelmemtok/sp50validoverallgroundingseo
mistralbest5.1 GB52.35.245s100%3.84.534
llama3.2:3b2.8 GB26.5213.96s100%3.84.133.87
phi34.0 GB75.273.796s73%3.453.823.36

Quality is the Anthropic LLM-judge (1–5). CPU-only Windows box (no GPU) — throughput is conservative/noisy.