Major Issues w/ the New Models
Hello, I love Venice and have been a Pro subscriber for a while. But the new models are simply not working. I’m referring to any of the models that include the word “Venice.” I’ve tried everything at this point to get them to work correctly, but I keep running into these things:
-They forget what I said just one or two messages ago.
-They fail to grasp the nuances of what I’m saying. Even things that are extremely easy.
-They aren’t including my custom instructions in responses.
-They get confused easily.
-They aren’t working as well with Characters.
I’ve tried disabling the Venice system prompt, cycling through the different new models to try to get one to work right, altering my custom instructions, etc., but it just seems like this is a huge downgrade.
I strongly (and I do mean strongly) recommend that Venice rolls back and includes the models that made them successful to begin with. For the new models that have “Venice” in them, they need rigorous testing and fixes before they are ready for prime time. And they might never be (which is fine). I’m not the only one who has run into issues with them.
Please don’t remove the previous models completely. Please.
Thanks!
Log in to comment and vote
Comments8
Blue Goose
Mar 16
Yeah, this is insane. I can’t say I have ever Loved Venice, it is not the great ChatGPT replacement that they advertise and if not for the lack of restrictions I probably wouldn’t bother paying for the subscription. But the last two weeks, it’s like Venice is actively going out of it’s way to ignore even the most basic instructions, sometimes just messing them up but most of the time just outright ignoring me. They need to fix this now, or I’m canceling my subscription when payment is next due.
Black Tree
Nov 18, 2025
Venice Ai is simply not working well for me on my iPad. I have problems typing on the keyboard, the space on the keyboard does not work, you do not allow us to speak the text, it is difficult to get to the creation page to find what you want to create… Just a lot of problems to be fixed on iPad before this is usable
Jack
Aug 6, 2025
We’ve made a number of improvements to our model lineup in recent weeks. From the legacy model list, Llama 3.1 405B is still available.
Sapphire Crab
May 7, 2025
To be honest, the new Dolphin-Mistral model is surprisingly unfiltered, but like you mentioned, it still can't hold a candle to the capabilities of the older models.
Sapphire Crab
May 7, 2025
I completely agree with you. The older Llama 405B, and even the 70B, were far superior to the new Venice model. And now that they've fixed the issues with the Llama 405B, I'm baffled as to why they wouldn't stick with it. I know it's considered an "old" model from last year, but honestly, it's still one of the best conversational models out there.
Jack
May 7, 2025
Thank you for sharing this detailed insight. Could you share some specific examples of these shortcomings, especially instances where the model forgets what you said just one or two messages earlier?
You can also report those conversations directly in the UI so we can take a closer look.
Rose Coaster
May 13, 2025
I've recently seen this too , the story generation was really good. It's still creative to a degree but if I'm creating an erotic story and give it instructions it's like it just wants to rush to the end and finish it. If I say redraft the story and don't rush, I will say ok then so the same again, then I'll say, why aren't you following the instructions it will apologise and do the same again even though I've given it a multi chapter overview and told it to write chapter 1 only. So it understands that I'm telling it to not end chapter 1 with sex but still does it. its also become a little less creative for me but might become better. I removed all my fixed prompts by the way from the settings. I've also noticed the forgetting. So if I've said at the start that James is with his girlfriend Sarah and then 6 prompts later say write a chapter about James and his girlfriends date, it will add a new person in forgetting about sarah. It could also be how I'm progressing the context though, I get that.
Jack
May 14, 2025
I’m guessing this is on Venice Uncensored? Could you share the temperature and top p values you are using? An example chat would be super helpful in tracking this down.