Architecture-first, efficiency-focused foundational LLMs engineered to deliver stronger reasoning per unit of compute. Enables previously impractical applications by unlocking compute-efficient reasoning and a unified foundation-model stack.