At a Glance · AI-generated

DeepSeek released V4.1-Flash on September 10, a new AI model featuring a 1 million token context window and optimized memory efficiency through FP4 key-value caching and cross-layer attention reuse. The company claims the model outperforms its flagship V4-Pro while significantly reducing memory requirements for AI agents. The architecture improvements enable practical deployment of long-context applications with lower computational overhead.