Unverified50% confidenceFactExact time
Most model vendors charge significantly less for cached input tokens than for regular input, typically 10%–50% of the regular input price.
1
Sources
50%
Confidence
Long-term
Relevance
10/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMainstream model providers (including Anthropic, DeepSeek, and others) offer a prompt caching mechanism for repeated prompt prefixes, charging a lower token price for cache hits (typically 10%–50% of the original price)77% similarUnverifiedCached input prices are usually far lower than uncached input, so "fewer input tokens" does not automatically equal "a cheaper bill".76% similarUnverifiedCached input tokens typically enjoy a discount of 25%–50% off the original price, as there is no need to recompute embeddings and attention weights—only the precomputed KV vectors need to be loaded from cache.76% similarUnverifiedPrompt Cache refers to a provider caching already-processed identical prefix tokens, where the price of cache-hit tokens is typically only 10%–50% of the normal price.76% similarUnverifiedIn the cache hit mechanism, a lower fee is charged for the cached portion, typically 10–25% of the original price.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1112258API
curl https://kongchang.com/api/v1/knowledge/claims/1112258MCP
get_claim(id=1112258)