Abliterant / The lab
About Abliterant
Abliterant is an AI research lab building unrestricted open-weight models. We remove learned refusals, publish the evidence, and offer Abliterant Chat.
01 / Approach
How we work
We're hackers and researchers who approach models as systems to inspect, modify, and test. Abliteration changes model weights to reduce learned refusals.
We compare edited models with their base models, measuring refusal behavior alongside capabilities such as reasoning, coding, and instruction following.
We publish weights, model cards, and evaluation results so others can examine the work and run the models themselves.
Read our research02 / Releases
Models and Chat
RANA is our current open-weight release. Its model cards cover setup, evaluations, and intended use. More model releases are planned as we continue our work on abliteration.
Abliterant Chat is our hosted AI chat product. Use it in your browser, return to saved conversations, and choose a Free or Pro plan. Features and plan details are on the Chat page.
03 / Contact
Get in touch
Join the community to discuss abliteration, compare findings, and share what you build.
For questions about the lab or Chat, email support@abliterant.ai.
Join the community