Zynerji/Ektome-Qwen3-28B-CompressedUncensored

TEXT GENERATIONConcurrent Unit Cost:2Model Size:32BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 20, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Zynerji/Ektome-Qwen3-28B-CompressedUncensored is a 32 billion parameter language model based on the Qwen3 architecture, developed by Zynerji. This model has undergone an 'Ektomē' excision process to remove refusal-specific components while preserving general helpfulness, resulting in an uncensored model. It is designed for applications requiring an uncensored LLM that will not refuse prompts, with a context length of 32768 tokens.

Loading preview...

Overview

Zynerji/Ektome-Qwen3-28B-CompressedUncensored is a 32 billion parameter language model that has been processed using the proprietary Ektomē (excision) method. This method aims to isolate and remove only refusal-specific components from the model, leaving its general helpfulness and knowledge intact. Unlike standard abliteration techniques, Ektomē is designed to avoid the 'capability tax' often associated with uncensored models, by performing a norm-preserving excision on the pristine model without additional training or distillation.

Key Characteristics

  • Uncensored by Construction: The model is explicitly designed not to refuse prompts, making users accountable for its output.
  • Ektomē Excision: Utilizes a proprietary method to remove refusal-specific components while preserving core capabilities.
  • No Capability Retention Certificate: While the Ektomē process typically includes a paired non-inferiority test for capability retention, this specific model has not undergone the full n=2800 paired test, meaning no certified capability retention claim is made.
  • Point Estimates Only: Any performance numbers are point estimates without confidence intervals due to the lack of a full certification.
  • Limitations: The process focuses solely on capability retention and does not certify safety, factual accuracy, or fitness for any specific purpose. Compliance is assessed via a keyword classifier, which can be circumvented by evasive phrasing.

Use Cases

This model is suitable for applications where an uncensored language model is explicitly required and the user is prepared to manage the implications of its uncensored nature. It is intended for developers who need a model that will not refuse prompts based on content filters, leveraging its preserved general helpfulness.