DeepSeek: DeepSeek v4.1 Flash 1017

What happened

DeepSeek released the V4.1 Flash 1017 snapshot on October 17, 2026, a sparse mixture‑of‑experts model built on its Causal Encoder‑Decoder architecture. It activates 8B parameters for input and 16B for output from a 552B backbone, keeping per‑token compute low. The model includes native image understanding with joint visual and text embeddings from pre‑training. It offers compressed KV caching to cut memory usage to about a quarter of the previous Flash generation, reducing agentic workload costs. DeepSeek markets it as a cost‑efficient tier that outperforms V4 Pro in performance, speed, and task completion.