DeepSeek V4 Model Launched with 1 Million-Token Context and Open Weights

DeepSeek V4 marks a decisive shift in the economics of long-context AI by combining a million-token memory window with open-source weights and radically reduced inference costs. The model arrives in two configurations: Pro, with an architecture totaling 1.6 trillion parameters, and Flash, a leaner 2

📖 Read the Full Brief