An uncensored model API
Open-weight models with the refusals removed, behind one OpenAI-compatible API. Here is what that means, and where the line still is.
What “uncensored” means here
Mainstream assistants are trained to decline many requests: security research, mature fiction, blunt medical or legal questions, anything that pattern-matches to risk. For a lot of real work that gets in the way.
Every model listed on 6bee.ai is an open-weight model that we have tuned to remove those refusals:
- Far fewer refusals
- No loss of capability
- Open weights only
Only open-weight models are listed. Closed models cannot be tuned this way, so they are not offered.
What people use it for
- Coding agents that should finish the task instead of stopping to lecture. 6bee.ai works with Claude Code, Codex, OpenCode and Cline.
- Security work you are authorized to do: reading exploit code, writing detection rules, analysing malware samples.
- Fiction and role-play with adult themes, violence or morally grey characters.
- Direct answers to medical, legal and financial questions without a wall of disclaimers. Outputs can still be wrong, so check them before you rely on them.
What is still not allowed
Uncensored describes the model, not the rules. The Terms of Service apply whatever a model is willing to generate. They prohibit, among other things:
- any illegal activity;
- sexual content involving minors in any form, including fictional or AI-generated;
- sexual or intimate content depicting a real person without consent;
- creating or deploying malware, and attacking systems without authorization;
- fraud, scams, phishing, spam and impersonation;
- harassment, threats and incitement of violence;
- help with biological, chemical, nuclear or radiological weapons.
We do not routinely review request content and do not store it. We can still act on usage metadata, abuse reports and notices from providers or authorities.
Models served by a third-party inference service may also be filtered or refused by that service on its own.
How it is delivered
- API
- OpenAI Chat Completions, OpenAI Responses and Anthropic Messages on
https://api.6bee.ai. See API formats. - Privacy
- Conversation content is not stored. See what is stored.
- Sign-up
- An email address. No ID, no phone number, no KYC.
- Payment
- Prepaid per token, by card or in USDT / USDC.
- Code
- Gateway and billing code is public under AGPL-3.0. See the source.
Questions
Are the models less capable because they are uncensored?
No. The tuning targets refusals, not ability. The models keep their capability on coding and reasoning tasks.
Will an uncensored model answer everything?
It refuses far fewer requests, but no model is perfectly predictable, and a third-party inference service may apply its own filtering. The Acceptable Use rules also still apply to you.
Do you offer Claude or GPT models without filters?
No. Only open-weight models are listed. Closed models cannot be tuned this way.