EldritchLabs/Aura-Prototype-26B-A4B

Hugging Face
VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:2Model Size:26BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 18, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

EldritchLabs/Aura-Prototype-26B-A4B is a 26 billion parameter merged language model, created by EldritchLabs using the experimental 'aura' merge method. This model combines several base models, including those focused on reasoning and fiction, to achieve a broad range of capabilities. It is designed for general language tasks with a context length of 32768 tokens, leveraging a unique merging process for its architecture.

Loading preview...

Aura Prototype 26B A4B: A Merged Language Model

EldritchLabs/Aura-Prototype-26B-A4B is a 26 billion parameter language model developed by EldritchLabs. It was created using mergekit with the experimental aura merge method, which features a live heatmap visualizer and took approximately 10 hours to process. This model integrates several pre-trained language models, including Gandalf69/Adversary-26B-A4B-v0, TheDrummer/Orion-26B-A4B-v1.1, Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2, and electroglyph/gemma4-26b-fiction-bf16.

Key Characteristics

  • Architecture: Based on Gemma4ForConditionalGeneration.
  • Merge Method: Utilizes the novel aura merge method, an experimental technique developed by EldritchLabs.
  • Component Models: Blends models with diverse focuses, including reasoning and fiction, suggesting a versatile capability set.
  • Configuration: The merge process involved specific parameters for weight, density, epsilon, and pinocchio settings for each component model, alongside advanced optimization parameters for the aura method.
  • Tokenizer: Employs a union tokenizer source for comprehensive vocabulary handling.

Good for

  • Exploring models created with advanced merging techniques.
  • Applications requiring a blend of reasoning and creative text generation, given its constituent models.
  • General language understanding and generation tasks with its 26B parameters and 32768 token context length.