That's good to hear. I dropped at the time they started training their own models, and it taking forever, while simultaneously finding success with imagegen, and kinda pivoting to that, which further delayed their text model, at which point I stopped kinda checking them out.

It definitely took forever, though I’m given to understand the training finished a while ago and they’ve been stuck due to lack of hardware.

Coincidentally however, just yesterday they announced that it arrived this week. The newest model, based on Llama 3.1, should be available for use next week.