Unverified50% confidenceFactExact time
Anthropic's input_tokens does not include cached tokens; total input equals cache_read_input_tokens + cache_creation_input_tokens + input_tokens, whereas OpenAI includes cached tokens in the input, so the two conventions differ and need normalization.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
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.78% similarUnverifiedPrompt Caching缓存的是输入而非输出，缓存LLM的输出几乎没有意义69% similarUnverifiedIn Anthropic's API billing system, prompt caching allows a stable prefix to be marked as cached content, and subsequent identical prefixes directly reuse the KV without recomputing attention, with these tokens recorded as "cache reads".68% similarUnverifiedThe prompt caching of many providers (such as Anthropic, OpenAI, and Google) uses token sequences as the unit of matching rather than strings, so subtle differences in spaces, line breaks, or encoding can cause cache misses.67% similarUnverifiedIn 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`.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1051988API
curl https://kongchang.com/api/v1/knowledge/claims/1051988MCP
get_claim(id=1051988)