Small Models Have Arrived

Small language models like GPT-5.6-Luna have become capable and cost-effective enough to enable new consumer and business applications, with inference costs dropping to pennies per request compared to dollars previously. This cost reduction removes a major barrier to building consumer AI products and could unlock significant demand for "fast/cheap/good-enough" models that handle routine, responsive tasks across businesses. The author argues that while frontier models remain essential for novel breakthroughs, the economics now favor deploying smaller models for the majority of work that requires responsiveness rather than genius-level problem-solving.