DeepSeek-V4.1-Flash-1017

Published 19 October 2026

DeepSeek-V4.1-Flash improves reasoning, coding, agentic, and visual performance while reducing execution time and output length. Cross-layer KV cache reuse and FP4 caching shrink its global KV cache footprint to 890 bytes per token, about one-eighth that of DeepSeek-V4-Flash.