Unverified50% confidenceFactExact time
Under the API billing model, about 65% of the equivalent bill comes from cache reads.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedFor agents continuously operating on the same codebase, cache reads can account for 70%–90% of total token consumption, making them the decisive factor in real costs.77% similarUnverifiedIn mainstream API pricing structures such as Anthropic's, the unit price of cache-read tokens is only 1/10 to 1/30 of standard input tokens.70% similarUnverifiedIn a typical agentic coding session, cache pricing (cache reads + cache writes) together accounts for about 56% of the bill, while input and output account for only about 44%.69% similarUnverifiedCached input reads enjoy a 90% discount (billed at 10% of the original price).69% similarUnverifiedPrompt caching is the highest-impact cost lever, with OpenAI's pricing sheet showing that cached inputs can save up to 90% in costs.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1043541API
curl https://kongchang.com/api/v1/knowledge/claims/1043541MCP
get_claim(id=1043541)