Verified65% confidenceFactExact time
DeepSeek-V4.1-Flash 的 KV 缓存压缩至每 Token 890 字节，仅为 V4 Flash 的四分之一
3
Sources
65%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
Unverified从V1到V4.1 Flash，单Token缓存从389000字节压缩到890字节，总共压缩了437倍78% similarUnverified经过压缩后，DeepSeek V4 Flash模型体积从接近500GB降至80.8GB74% similarUnverifiedDeepSeek V4 Flash的UD-Q8_K_XL量化版本文件体积为162GB，采用MXFP4+BF16混合精度，达到完全无损73% similarVerifiedV4 Flash的缓存大小只有V3.2的7%73% similarUnverifiedDeepSeek V4 Flash是主打高吞吐的云端模型，推理速度可达约每秒140 tokens73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924480API
curl https://kongchang.com/api/v1/knowledge/claims/924480MCP
get_claim(id=924480)