Dogoo3/Aletheia-12B

TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:Jan 18, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Aletheia-12B is a 12 billion parameter language model developed by Dogoo3, created through a merge of several existing models including MN-HyperNovaIrix-12B, Violet-Lyra-Gutenberg-v2, AngelSlayer-12B, FusionEngine-12B, and Patricide-12B. This model was specifically designed as an experimental attempt to achieve a more intelligent and creative 12B model, offering a 32768 token context length. Its primary use case is for creative text generation and general intelligent conversational tasks, with a focus on enhancing imaginative outputs.

Loading preview...

Aletheia-12B Overview

Aletheia-12B is a 12 billion parameter language model developed by Dogoo3, resulting from an experimental merge of five distinct models: MN-HyperNovaIrix-12B, Violet-Lyra-Gutenberg-v2, AngelSlayer-12B-Unslop-Mell-RPMax-DARKNESS, yamatazen/FusionEngine-12B-Lorablated, and redrix/patricide-12B-Unslop-Mell. This merge was conducted using mergekit with the goal of creating a more intelligent and creative model for personal use, which has now been released to the community.

Key Capabilities

  • Enhanced Creativity: Designed to produce more imaginative and creative text outputs.
  • Intelligent Responses: Aims for higher intelligence in conversational and generative tasks.
  • Flexible Merging: Built from multiple base models, allowing for diverse underlying strengths.
  • Extended Context: Supports a context length of 32768 tokens, suitable for longer interactions.

Good For

  • Creative Writing: Ideal for generating imaginative stories, scripts, or descriptive text.
  • Experimental Merges: Serves as a base or component for further model merging experiments.
  • Personalized AI Applications: Suitable for users seeking a highly customizable and creative 12B model.
  • General Conversational AI: Can be used for various dialogue-based applications where creativity is valued.

Known Issues

  • May exhibit occasional refusals.
  • Can sometimes be repetitive in generated text.
  • May experience issues with user role retention (talking as {user}) after approximately 10,000 context tokens.

Finetunes
1 models