[DanielVZ]

I’m pretty sure they were using a high cost LLM to summarize, and for embeddings you only need Ada, which is orders pf magnitude cheaper.