How get cached Input tokens?

[JulioVianaDev]

Someone knows how i can get the cached inputs? Into usage of ai i receive the output, the input and total tokens but i not know how much of this tokens are cached

[barry166]

From the current types/adapters, the field you want is usage.promptTokensDetails?.cachedTokens.

promptTokens is the total input token count; cache reads are the optional breakdown:

for await (const chunk of chat({ /* ... */ })) {
  if ("finishReason" in chunk) {
    const cachedInputTokens = chunk.usage?.promptTokensDetails?.cachedTokens ?? 0
    const cacheWriteTokens = chunk.usage?.promptTokensDetails?.cacheWriteTokens ?? 0
  }
}

For the OpenAI adapters, TanStack AI maps provider values like input_tokens_details.cached_tokens / prompt_tokens_details.cached_tokens into cachedTokens. If that property is undefined, the provider or model did not report cache usage for that request, so there is no separate cached-token number to read from the normalized usage object.