Motif 3

Motif 3 is Motif Technologies’ large open Mixture-of-Experts model with 314 billion total parameters and 13.2 billion active parameters per token. Its model card describes 384 routed experts, eight active experts plus a shared expert, and a native 262,144-token context. The design targets agents, tool use, coding, mathematics, and multilingual work, with particular emphasis on Korean and legal and financial text.

The small active count does not make Motif 3 a small model. Its 314 billion total parameters still determine the weight and serving burden even though only part of the experts compute for each token. The official card documents vLLM and separate quantized checkpoints; the MIT license applies to the published weights.