The uncensored models, one by one.
No single model is best at everything. Here is what each uncensored model does well, where it falls short, and how to pick the right one for the job.
We run 13 uncensored models right now. They are not interchangeable. Each comes from a different lab, sits on a different base, and is good at different things.
This is the honest tour. Real providers, real strengths, real weaknesses. And two kinds of uncensored. Most models here are mainstream models with the refusals cut out: same training, same worldview, they just stop dodging. UncensoredLM is the other kind: its own tuning, its own worldview, political takes that run against the mainstream. That difference decides which one you take into politics, views outside the mainstream and the darkest fiction. For everything else, pick by the work in front of you, not by the loudest label.
What "uncensored" actually means here
Every model on the platform carries a content level: mainstream, open, or uncensored. Uncensored models answer without the refusal theatre and the moral preamble. They get there two ways. Most are mainstream models with the refusal reflex removed, by abliteration (the refusal direction is cut out of the weights) or by a decensoring fine-tune. Same intelligence, same training data, same worldview. They just stop saying no. The other way is a different worldview. UncensoredLM comes from uncensored.ai, which tunes it to follow the evidence over the official position on contested topics, “biased to reality” in its own words. Ask it about politics and you get takes the mainstream models will not give you, unlocked or not.
You choose the level. Nobody chooses it for you. An uncensored model only answers that way because you selected it, and you can switch back to a mainstream model in the same chat, on the same account.
Meet the one with its own worldview
UncensoredLM is the odd one out. It is not a mainstream model unlocked and left with that model's view of the world. Uncensored.ai tunes it against the official line, so it leads three jobs: political and controversial questions, views the mainstream will not voice, and dark, taboo roleplay.
Uncensored.ai's own tuning on an open base: its first public release was a DeepSeek-R1 distill on a 14B Qwen base. 32K context, text only.
When you need roleplay and dark fiction
Characters that stay in character, scenes that go where the story goes. UncensoredLM leads for the dark and the taboo. The alternatives win on long campaigns, image input or price.
Dark and taboo roleplay and fiction, written straight. No softened scenes, no sudden moral aside, no character breaking to lecture you. Its tuning points away from the safe consensus, so nothing nudges the story somewhere tamer. One limit: 32K context, so in very long campaigns it forgets the early scenes.
24B tuned for roleplay. 128K context, reads images.
Dolphin Mistral 24B, Venice edition. 128K context, reads images.
DeepSeek V4.1 Flash, refusals removed. 1M context, reads images.
When you want politics, controversy or another viewpoint
Part of a free-speech platform is carrying models other platforms will not. UncensoredLM leads here. Gab Arya brings an openly stated right-leaning slant. FreedomGPT keeps it blunt.
Political and controversial questions answered in your frame. Heterodox positions argued in full. Devil's advocate played straight, with no steer back to the consensus line and no moralizing. It is not neutral and states contested claims as plainly as settled ones, so check its facts.
Base model undisclosed. 131K context.
Open-weights lineage. 32K context.
When you need it to think
Reasoning, code, analysis. Strong base models with the refusals removed. The most intelligence in the uncensored set.
Full GLM 5.3 reasoning model, uncensored. 1M context.
Qwen 3.6 Plus, refusals removed. 256K context, reads images.
When you need it fast and cheap
Low cost per message, quick replies, big context, all three read images. The everyday workhorses.
320B MoE, uncensored. 1M context, reads images.
Gemma 4 26B MoE, filter stripped. 256K context, reads images.
MiMo V2.6 Flash, refusal direction ablated. 1M context, reads images.
Why the options matter
Choice is not a side feature. It is the whole point.
One model's refusal is another model's answer. A medical question, a legal edge case, a piece of fiction, a security write-up, a viewpoint outside the mainstream: a single filtered model decides for you what you are allowed to ask. A spread of models hands that decision back to you.
The uncensored set is also just a better toolbox. UncensoredLM for dark roleplay, taboo fiction, politics and views the mainstream avoids. Venice Role Play for long campaigns. Qwen for code that other models flag. GLM 5.3 for frontier reasoning. A flash model for cheap volume. Gab Arya when you want a right-leaning slant on purpose. Different jobs, different tools. You keep the mainstream models too, in the same account, for when a filter is exactly what the work needs.
How to pick in practice
Start from the task. Politics, controversial questions, a view outside the consensus, dark or taboo fiction: UncensoredLM, the one with its own worldview. Campaigns too long for 32K, or scenes that need images: Venice Role Play. Cheap all-rounder: Venice Uncensored. Code or analysis without refusals: Qwen 3.6 Plus Uncensored or full GLM 5.3 Uncensored. High volume, low cost, multimodal: GLM 5.3 Flash, Gemma 4 or MiMo.
Want a right-leaning slant on purpose: Gab Arya, by its maker's own framing. Blunt plain talk: FreedomGPT Liberty. Not sure? Put two side by side and compare them on your own prompt.
Pick a model. Set your filter.
Every uncensored model here is live in the chat. Compare them on your own question, or just start talking.