DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression

Name: DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
Uploaded: 2026-10-21T04:11:00.397550+00:00
Duration: 558 s
Description: DeepSeek-V4.1-Flash is a 552B-parameter multimodal MoE model that cuts global KV cache to 890 bytes per token, about a quarter of its predecessor's, while improving performance. It combines a 20-layer causal encoder with a 20-layer decoder, compressed sparse attention with cross-layer KV reuse, FP4 main-KV quantization, and bounded-replay SWA recovery, reaching 74.2% on DeepSWE v1.1.

DeepSeek-V4.1-Flash is a 552B-parameter multimodal MoE model that cuts global KV cache to 890 bytes per token, about a quarter of its predecessor's, while improving performance. It combines a 20-layer causal encoder with a 20-layer decoder, compressed sparse attention with cross-layer KV reuse, FP4 main-KV quantization, and bounded-replay SWA recovery, reaching 74.2% on DeepSWE v1.1.

Open the full episode