>>715699946
throw a tenner on openrouter and you can test different LLMs. opus is great, but too expensive. I've been using sonnet 3.7 with github. i don't use sillytavern. I use risuai with korean assetbots though. gemini 2.5 pro is pretty decent and is coherent and remembers everything well. deepseek isn't really good at anything. it's 'creative' as in it goes off the rails a lot. claude provides more fun character interactions, but is bad at remembering time, events in history, and registering physical space correctly in a narrative. Korean bots have have built in "affection" meters, some even have music and built in animations for stuff. I am not korean btw I just used google translate to figure it out.

>>715703124
what model are you using? gemini works fine for that sort of thing.

>>715699946
throw a tenner on openrouter and you can test different LLMs. opus is great, but too expensive. I've been using sonnet 3.7 with github. i don't use sillytavern. I use risuai with korean assetbots though. gemini 2.5 pro is pretty decent and is coherent and remembers everything well. deepseek isn't really good at anything. it's 'creative' as in it goes off the rails a lot. claude provides more fun character interactions, but is bad at remembering time, events in history, and registering physical space correctly in a narrative. Korean bots have have built in "affection" meters, some even have music and built in animations for stuff. I am not korean btw I just used google translate to figure it out.

>>715703124
what model are you using? gemini works fine for that sort of thing.