“DeepSeek-V4-Flash is a Mixture-of-Experts (MoE) language model with 284 billion total parameters and 13 billion activated parameters.”