VercelAI instrumentation missing cached token usage data

[raymondhechen]

Is there an existing issue for this?

• I have checked for existing issues https://github.com/getsentry/sentry-javascript/issues
• I have reviewed the documentation https://docs.sentry.io/
• I am using the latest SDK release https://github.com/getsentry/sentry-javascript/releases

How do you use Sentry?

Sentry Saas (sentry.io)

Which SDK are you using?

@sentry/node

SDK Version

10.17.0

Steps to Reproduce

Using the default VercelAI instrumentation plugin with streamText method from AI SDK. Also using Anthropic and AWS Bedrock as model providers. Here's a screenshot of the setup in the Sentry webapp:

Expected Result

Expected option to segment by cached input tokens in Sentry.

Actual Result

Option to sum/show cached input tokens doesn't exist. (I tried searching for the option too)

[linear]

JS-1035

[Lms24]

Hey @raymondhechen thanks for writing in!
To confirm: Are you missing data around cached input tokens (vs. generally input tokens) you'd expect the Sentry SDK to send?

I'm going to ask @RulaKhaled to further triage and assess this.

Edit: I tentatively removed the Bug label in favour of Improvement because I believe this is rather about adding data on top of our offering. Feel free to correct me on that 😅

[RulaKhaled]

hello @raymondhechen, do you see gen_ai.usage.input_tokens.cached in the span attributes though?