With the speed they're adding features shows you're correct.

Which others?

Well, they do need to try to compete with SubQ’s longer context and image processing that has been coming out ahead in benchmarks, plus the thousand self-hosted options that now exist.

I used the model and it was good enough for email classification. I was wondering if I'm supposed to be doing multiple runs for topic classification with grouping related topics i.e. if its sport,economy or tech related news and then another run for soccer,basketball or something else or batching them like the News article topic classification example in a single run.

I don't think they can. But they probably monitor usage and test for patterns to then block/ban accounts

Curious why you use Gemini 3.5 Flash Lite instead of SAM/Moondream? Any specific workflows where it outperforms?

Judging from the tilte, I guess dreams.

Automatic tagging of images is what I'm doing with it. I'm using a modified Gemma model, which ingests images (and audio, for that matter).

it's actually more usefull than it looks. it is not much useful on its own but it can play a big role in a complex workflow or custom harness.

i used luna 5.6 as the ranker/filter to remove useless items from upstream rag(mixing embedding, regex etc) and it did a good job but took 5-10 seconds each time. now less than 1 second and no observable precision loss. hope it helps