Unverified50% confidenceFactExact time
With the cached input mechanism, if a team can design a stable, reusable context prefix, the actual input cost can drop to one-twentieth of the original.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCache hit discounts target repeated system prompts or context prefixes: if this content has already been cached, the provider does not need to recompute it and can heavily discount the cost of the hit portion, commonly to 10% to 25% of the original price.70% 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.69% similarUnverifiedIn frameworks where prompt caching is already enabled, the unit price for cache hits is only 1/50 that of new input, so cost savings are limited when the volume of judgments is not large.68% similarUnverifiedOverly long system prompts bring hidden overhead: for example, when the system prompt has 500 tokens while the actual input has only 50 tokens, it means 91% of the cost goes to boilerplate text; auditing and trimming the system prompt by 40% can reduce costs.67% similarUnverifiedIn the cache hit mechanism, a lower fee is charged for the cached portion, typically 10–25% of the original price.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1038766API
curl https://kongchang.com/api/v1/knowledge/claims/1038766MCP
get_claim(id=1038766)