# DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression > Source: > Published: 2026-09-18 02:30:24+00:00 The widespread adoption of long-horizon agents has made model workloads increasingly input-heavy. Although prior work has substantially reduced the cost of long-context computation, prefill remains computationally expensive, and large KV caches continue to strain HBM and SSD capacity and data-transf