Thinking Machines Inkling 2 Small vs. Baidu ERNIE 5.5 Flash Base

In November 2026, both Thinking Machines and Baidu introduced significant Mixture of Experts (MoE) models. Thinking Machines released Inkling 2 Small with 594B total and 17B active parameters, while Baidu launched ERNIE 5.5 Flash Base with 280B total and 14B active parameters. Reported facts show distinct scaling strategies in balancing total parameter capacity against active inference footprints. Analytical review indicates both models prioritize efficient active parameter utilization relative to their massive backbone sizes. Hands-on evaluation was not conducted.