ACE-Step/Ace-Step1.5

Text-to-Audio • Updated Feb 3 • 39.9k • 679

Note Local Music Generation. Actually very good for a local model. It's far from perfect, it's not Suno. It's more like "we have Suno at home", but still, you can make pretty dope music with it.

ReadyArt/Omega-Evolution-27B-v2.0

27B • Updated 11 days ago • 33 • 11

Note Feels really "early alpha". More censored and less coherent than BlueStar 2. Probably undercooked. Function calling and thinking work very well, though.

maldv/Loqwqtus2.5-32B-Instruct

Text Generation • 33B • Updated Mar 19, 2025 • 8 • • 2

Note Qwen3.2 -> fine-tune EVA -> fine-tune CoT model. Tested quickly, seems okay. ChatML + Think tags

SicariusSicariiStuff/Impish_Magic_24B

24B • Updated Nov 10, 2025 • 12 • 45

Note Pretty generic Mistral 24B RP model. Not great, not bad. Just meh.

LatitudeGames/Harbinger-24B

24B • Updated May 16, 2025 • 758 • 80

Note MS24B 3.1 (Base) fine-tune. ChatML. Trained on narrative at the second person *you do this*. Sounds good. Very versatile but sadly a bit too "dumb" to be featured on other listings.

google/gemma-3-27b-it

Image-Text-to-Text • 27B • Updated Mar 21, 2025 • 879k • • 1.95k

Note Meh. Extremely memory hungry at 16K context, unless you quant the KV cache which comes with its host of issues. It's to a point it's not worth testing for me (and neither are any finetunes by extension)