Firebase AI - no support for usageMetaData

[kareemayoub]

The flutter SDK does not include usageMetaData for the Gemini 2.5 live api , but it includes it only for Flash or Pro models. This is a critical feature for an app we are currnently working on to properly calculate the usage metadata data and related costs for each session.

[ushxyuki]

I agree this would be useful, especially for production apps that need per-session usage tracking.

usageMetadata is important because token counting before the request is not enough for accurate cost or quota tracking. A preflight token count can estimate the prompt size, but it does not fully replace the final response metadata, especially when generated output tokens, thinking tokens, cached content, or streamed responses are involved.

It would be helpful if the Flutter Firebase AI SDK exposed usage metadata consistently across supported Gemini 2.5 models and APIs, including fields such as:

• Prompt token count
• Candidate/output token count
• Thinking token count, when applicable
• Total token count
• Cached content token count, when applicable

A temporary workaround could be routing requests through a backend using the Gemini or Vertex AI API directly and returning the usage metadata with the response, but that removes some of the convenience and security benefits of using Firebase AI Logic directly from Flutter.

Native support in the Flutter SDK would make usage analytics, cost estimation, and quota monitoring much cleaner.