I have an app, glutenornot (https://github.com/amr05008/glutenornot.com), that currently uses opus 4.8 to classify and flag gluten containing ingredients from bar code scans, ingredient labels and restaurant menus.

I have an app, glutenornot (https://github.com/amr05008/glutenornot.com), that currently uses opus 4.8 to classify and flag gluten containing ingredients from bar code scans, ingredient labels and restaurant menus. I tested jev against the current production setup against 1000 food labels from open food facts across multiple languages. result was pretty incredible tbh. equaled opus on flagging items with gluten , came in at $0.03 for jev vs $8.45 for opus 4.8 , median latency with jev was .17s vs 2.89 seconds for opus 4.8 , will test some more but def thinking about swapping my app to use this now as primary model. this experiment (and some others) here: https://github.com/amr05008/jev-sandbox