Model
158d agoDeepSeek V3: New Mixture of Experts Model
DeepSeek V3 launches a groundbreaking mixture‑of‑experts architecture that scales language models without linear growth in parameters. By routing queries to specialized experts and using a novel SwiGLU activation, the model delivers state‑of‑the‑art performance while reducing inference cost.
The Signal
Its open‑source API invites researchers worldwide to experiment and accelerate innovation for collaboration.