DeepSeek: DeepSeek v4.1 Flash 1017

DeepSeek released V4.1 Flash 1017 on October 17, 2026, a sparse mixture-of-experts model built on the company's new Causal Encoder-Decoder architecture that activates only 8B parameters on input and 16B on output from a 552B-parameter backbone. The model features native image understanding, compressed KV caching that reduces memory costs to roughly a quarter of the previous generation, and is optimized for coding, agentic tasks, and long-context analysis at a cost-efficient $0.15/$0.60 per 1M tokens. DeepSeek reports that V4.1 Flash exceeds its V4 Pro predecessor in performance, speed, and task completion time.