Comment by maestroquirk

We've been running a small finetuned VLM for OCR and yeah... cost is like 1/3rd of the usual vision APIs. Small models are kinda the whole product for us