Unverified50% confidenceFactExact time
In Anthropic's usage object, `input_tokens` counts only the portion after the last cache breakpoint, and the complete input is the sum of the three fields `input_tokens`, `cache_creation_input_tokens`, and `cache_read_input_tokens`.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedCache behavior on Anthropic can be verified via the cache_read_input_tokens and cache_creation_input_tokens fields; if both are 0, nothing was cached.74% similarUnverifiedIf most of the prompt is repeated and fixed while cache_read_input_tokens stays near zero for a long time, it can be almost certain that the prefix changed between requests.67% similarUnverifiedThe output vocabulary size of a pointer network is dynamically tied to the input length, and each output token corresponds to an index in the input rather than being generated from a fixed vocabulary.66% similarUnverifiedWhether prompt caching hits depends on byte-for-byte consistency of the request prefix; if even one character at the beginning differs, the entire cache is invalidated.64% similarUnverifiedFor text classification tasks, only the output vector of the last token is needed for classification, because it can attend to all preceding context63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1049114API
curl https://kongchang.com/api/v1/knowledge/claims/1049114MCP
get_claim(id=1049114)