Not exactly “serverless” but I built something similar with AWS SageMaker, which has elastic inference abilities. it’s rather fast to spin up and down.

Also, when it comes to summarization- you don’t really need to infer each run, you can throw up a pretty simple caching system. Which means repeat requests are far cheaper and faster.

I used cloudflare workers as a proxy / caching layer with KV in front of an AWS lambda to do article extraction and SageMaker spinup (with a small cache on the AWS side too- to catch in progress jobs)

This is v2 of the service that I launched last year. I am not aware of any ML serverless offering that does text summarization.