Update time!
Deepseek v3 is really good! Can't say it's better than gemini's current pro model, or OAI's (I stopped paying for openai, too expensive when I was getting nearly the same results using google's AI for free) but it's definitely getting up there, especially when compared to the filters it uses.
Gemini will cut itself off if it detects "extreme" content, but deepseek will only, either, disengage or continue. Which is great, because you can easily cheese it by having it start with a phrase of an agreement to continue because, and this is my favorite part, it's REALLY good at following directions you give it.
It's a solid 8/10 model, and if you're decent at fiddling with prompts you can make it say and do some really extreme stuff with shockingly good results.
Not that it'll ever be actually useful for most people, but it's also open source, so you to can just download the whole thing and use it using a normal super computer (Seriously, it's a 600b MoE model, I doubt even the most hardcore of AI enthusiast have that type of power to run it, but it's really cool seeing these models offered to the public) and has an API you can use for $1.30 per million tokens (Though, they have a discount going on where you can use it for about 0.40 cents per million tokens up until February 08). And if you just wanna dick around with it, they have a chat mode you can use for free on there website, downside is you can't tune the temp or set the context length, but I haven't had much context loss issue yet.
Just turn off "deepthink" if you use the free chat, it's mostly useless.
Deepseek v3 is really good! Can't say it's better than gemini's current pro model, or OAI's (I stopped paying for openai, too expensive when I was getting nearly the same results using google's AI for free) but it's definitely getting up there, especially when compared to the filters it uses.
Gemini will cut itself off if it detects "extreme" content, but deepseek will only, either, disengage or continue. Which is great, because you can easily cheese it by having it start with a phrase of an agreement to continue because, and this is my favorite part, it's REALLY good at following directions you give it.
It's a solid 8/10 model, and if you're decent at fiddling with prompts you can make it say and do some really extreme stuff with shockingly good results.
Not that it'll ever be actually useful for most people, but it's also open source, so you to can just download the whole thing and use it using a normal super computer (Seriously, it's a 600b MoE model, I doubt even the most hardcore of AI enthusiast have that type of power to run it, but it's really cool seeing these models offered to the public) and has an API you can use for $1.30 per million tokens (Though, they have a discount going on where you can use it for about 0.40 cents per million tokens up until February 08). And if you just wanna dick around with it, they have a chat mode you can use for free on there website, downside is you can't tune the temp or set the context length, but I haven't had much context loss issue yet.
Just turn off "deepthink" if you use the free chat, it's mostly useless.
