Skip to main content

Add Hermes 4 405b as a new Venice Large chat model, move Qwen3 235b to Venice Reasoning

Feature RequestsNew ModelChat

https://huggingface.co/NousResearch/Hermes-4-405B

Excells at steerability, is approximately twice as uncensored as Qwen3 according to Nous’ benchmarks, would be a good large-sized uncensored model to complement the venice-uncensored model, would replace the retiring llama 405b for character usage, and wins points for the cool factor

1 comment

Log in to comment and vote

Comments1

  • Rose Milkyway

    •

    Nov 1, 2025

    I used to follow Nous in the past. Their L3 finetune was decent. But it’s important to note, as I’ve learned from speaking to the developers, they define “steerability” kinda the opposite of what the word actually means. It was just better than L3 at adhering to the system prompt, but was in fact harder to steer in the user context than the base model and that was in fact the intention of Nous. It did allow suppressing censorship more reliably in the system prompt that way, but created some rigidity problems too as you had to be explicit about everything it is allowed in advance, even aside from censorship.

    Then they launched their own AI chat site and overcensored it into oblivion, threatening account suspensions for incessant false-positive detections. That left a bad taste about them in my mouth.