Overview
Both Thinking Machines and Baidu have released large-scale Mixture-of-Experts (MoE) models designed to optimize processing efficiency by activating only a fraction of their total parameters during inference.
Reported Architecture and Scale
*Analysis:* While direct benchmark comparisons are absent from the immediate evidence, the structural divergence between a 594B/17B configuration and a 280B/14B configuration highlights varying design choices in balancing capacity against memory footprint.