use case

Uncensored AI for roleplay and fiction

Creative work is where refusal behaviour is most visible: a scene that should be tense, ambiguous or dark gets flattened into a disclaimer. An uncensored model gives you more room, but voice and consistency still decide whether the output is usable.

Last updated: 2026-10-23

Pick for voice, not just willingness

Two models can both comply and sound completely different. Test a scene with dialogue, subtext and a change of emotion. Look for sentence rhythm, character consistency and the ability to hold a tone without repeating itself. A model that never refuses but writes flat prose is not a good fiction model.

Build a small state block

Long sessions fail when the model forgets who knows what. Keep a compact state block in the system message: characters, current location, open conflicts and facts that must remain true. Summarise older turns instead of resending the whole transcript. This also controls cost.

Separate the jobs

Mixing models is normal. One API key and one balance can cover all three.

Uncensored models to test

ModelTypeContextInputOutput
Magnum V4 72B
magnum-v4-72b
uncensored32.8K$12$20
DeepSeek V4.1 Flash 1017 Uncensored
deepseek-v4.1-flash-uncensored
uncensored1.05M$0.6$2.4
Weaver (Alpha)
weaver
uncensored8.0K$2$3
Cydonia 24B V4.1
cydonia-24b-v4.1
uncensored131.1K$1.2$2
Skyfall 36B V2
skyfall-36b-v2
uncensored32.8K$2.2$3.2
UnslopNemo 12B
unslopnemo-12b
uncensored1.02M$1.6$1.6
ReMM SLERP 13B
remm-slerp-l2-13b
uncensored6.1K$1.8$2.6
GLM-5.3 Flash Uncensored
glm-5.3-flash-uncensored
uncensored1.05M$0.165$0.618

Cost

Fiction sessions are input-heavy: the model rereads context on every turn. Keep history trimmed, use a smaller model for routine turns and reserve the larger model for scenes that need it. The pricing page has a request calculator; a long session can be many requests.

Boundaries

Fewer refusals do not remove legal and ethical limits. Sexual content involving minors, illegal activity and harassment are prohibited on every model. Read the acceptable use policy. If you publish the output, you are responsible for it.

Related

Read the abliterated models guide or browse the full uncensored catalog.

Questions

Which uncensored model is best for roleplay?

Test two or three on the same scene. Prioritise voice, character consistency and context handling over refusal rate alone.

How do I keep a character consistent?

Use a compact state block, summarise older turns and keep the character's voice examples in the system message.

Can I publish the output?

You are responsible for what you publish. Follow the acceptable use policy and the law where you operate.

Create account - no emailSee pricingBrowse models