[v5 and v4 ] StreamText does not have raw usage

[zhihengGet]

Description

generateText have usageMetadata but streamText does not, i need it for gemini models, i would like to have the usageMetadata in response.body . generateText seem to have it

ai/packages/google/src/google-generative-ai-language-model.ts

Line 284 in 26e7067

rawResponse: { headers: responseHeaders, body: rawResponse },

howeer, streamText does not have it :( for both v5 and ai@latest , i know v5 included more token type in the usage field but i want to access token count details audio..etc

AI SDK Version

ai@alpha and ai@latest

[dszczepaniak-cksource]

@lgrammel @dancer is there any possibility to access the response body object for the streamText method like for generateText? I have checked the documentation, and for streamText, I don't see a body field in the response object. My specific case is that I need it to access the usageMetadata field for Gemini calls to get the cached token count. It seems it is not possible to obtain this value when using streams.

Also, I wonder if this metadata field should be available in providerMetadata (currently it is not, or at least I don't receive it)?

[lgrammel]

There is no "body" per se, it is a stream. In ai sdk 5 you can access the raw chunks (enable setting on streamText then look for raw chunks).

example: https://github.com/vercel/ai/blob/v5/examples/ai-core/src/stream-text/raw-chunks.ts

[dszczepaniak-cksource]

It seems in V5 the usage object includes what I need:

{
  inputTokens: 3120,
  outputTokens: 440,
  totalTokens: 3560,
  reasoningTokens: 0, 
  cachedInputTokens: 0
}

Thanks!