@Pokee_AI Not open sourcing a 28b model, probably based off Qwen 3.6 27b anyways, and making it an API gotta be the greediest thing I have ever seen. Who the fuck cares if it has 10 million context if it can't even make proper code edits past 16k context.
@daniel_mac8 For anything other than basic work, Luna is simply not good, because it isn't as "wise" as those bigger models like Sol. Also output tends to be on the token heavier side, I guess it could be still a decent work horse model?
@eric_alcaide This way, more people with consumer gpu's can utilize the model at like ~95% quality, with faster t/s than something like a normal 4b quant or something.