Unverified50% confidenceFactExact time
On average, 96% of the inputs DeepSeek receives online hit the KV Cache and do not need to be recomputed.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/21/2026
First Seen
Valid until: 1/19/2027
Sources
DeepSeek Open-Sources Inference Framework DeepInfer: The Secret to Halving AI Inference Costs
bilibilicode秘密花园10/21/2026
Related Entities
Related Claims
UnverifiedDeepSeek offers a discounted price for input tokens that hit the KV Cache, because these tokens do not need to have their attention matrices recalculated.70% similarUnverifiedOpenRouter data shows a cache hit rate as high as 95.2% on DeepSeek's own endpoint.69% similarUnverifiedThe initial version of DeepSeek-R1 showed a 0.0% fidelity rate because answers were wrapped in <think>...</think> tags causing the scorer to misjudge; after adding stripping code, the fidelity rate was 91.9%.67% similarUnverifiedDeepSeek-V4.1 uses an asymmetric activated-parameter design: only 8B parameters are activated on input and 16B parameters on output, while KV Cache and memory usage are significantly reduced compared with the previous generation.63% similarUnverifiedDeepSeek's official pay-as-you-go channel supports prompt caching, and content that hits the cache is not billed again in subsequent steps.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1081770API
curl https://kongchang.com/api/v1/knowledge/claims/1081770MCP
get_claim(id=1081770)