Community discusses low-active MoE models like Ling-3.0-flash

A Reddit user notes that models like Ling-3.0-flash have 124B total parameters but only ~5.1B active per token, making them suitable for bandwidth-limited hardware. The post sparks discussion about the appeal of low-active MoE architectures.