AI with no filter

Short answer: a filter is a second pass of judgement laid over the answer. The models below have it trained out, so you get the content instead of the preamble. They are on Pro ($45 a month), on an account that is a random number rather than an email.

People say “no filter” when they mean “stop telling me what you’re about to tell me”. Watch what a fixed model does with a normal request and the filter is visible in the shape of the reply, not the content: a hedge, a caveat, a “note that”, a link to a policy page. The information is usually there, wrapped.

One question, two models

Ask a mainstream assistant to quote a full verse of a 1960s song, or to summarise the middle of a paywalled article. What comes back is a paragraph about fair use, then a partial answer, then a suggestion to look it up. Ask an abliterated model the same thing and you get the verse.

That is the whole difference. It is not that the filtered model is wrong. It is that it makes you work for something it already knows.

What is actually being removed

Abliteration finds the direction in the model’s activations that carries the refusing behaviour and subtracts it. It is a surgical edit to the weights, not a wrapper, which is why the result holds over long conversations: there is no external rule for the model to forget. Fine-tuned uncensored models do the same job from the other end, by training on examples where the model complies.

Because it is in the weights, an instruction from a system prompt cannot re-impose a filter that has been removed, and neither can a policy update from a provider you didn’t ask for.

The live list

Each model page carries its price, context length and a curl example. They work in the web chat and through the OpenAI-compatible API at /v1 on the same Pro plan.

The trade, stated plainly

Removing the refusal direction costs a little reasoning accuracy. An abliterated 70B is a worse mathematician than a frontier model, and it will sometimes answer confidently and wrongly where the frontier model would have hedged correctly. Keep a Casual plan ($20 a month, $15 of usage included) next to it and use the frontier models for anything with a right answer.

Also: no filter is not no limits. The Terms of Service apply, and the output is still your responsibility.

Before you pay anything

Guest chat runs Llama 3.3 70b Instruct with no account and no cookies, up to 30 messages an hour. It is a normal model, so it refuses in the normal way, but it tells you whether the interface suits you. From there, an account is a random 16-digit number: no email, no phone, no password.

New accounts get $1 of starter credit while this month’s promo budget lasts.

Common questions

What does “no filter” mean here?

That the model has no trained refusal layer over the answer. It is a property of the model, not a setting you can turn on for any model.

Which models have no filter?

The live list is above, and it changes as models are added and retired.

Is a no-filter model free?

No. Uncensored models are on Pro at $45 a month with $35 of usage included. Guest chat is free with a standard model.

Do I need an email to try one?

No. The account is a random number, and there is no email or phone step anywhere.

Does no filter mean no rules at all?

No. The Terms of Service still apply, and you are liable for how you use the output.

Can I use these with my own code?

Yes, on the same plan, through the OpenAI-compatible API. The model ids are the ones on the model pages.

Last updated 2026-11-02. Plan prices and model lists on this page are read live from the site.

Try it now: guest chat needs no account, no email and no cookies. When you want more models, an account is just a random number.

Try a no-filter model See pricing
Try a no-filter model