Sorihon/Celestial-Order-24B-2.75

TEXT GENERATIONPricing:Input $0.7 / Cached $0.04 / Output $1.16Concurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026Architecture:Transformer Featherless Exclusive Cold

Celestial-Order-24B-2.75 is a 24 billion parameter language model developed by Sorihon, created through a DARE TIES merge of pre-trained models, including Sorihon/CYDR-24B. This model is a refined iteration, building upon Celestial-Order-V2.5, and is designed for general language understanding and generation tasks with a context length of 32768 tokens.

Loading preview...

Model Overview

Celestial-Order-24B-2.75 is a 24 billion parameter language model developed by Sorihon. It is a merged model, specifically created using the DARE TIES merge method, which combines the strengths of multiple pre-trained language models. This version, V2.75, builds upon a previous iteration, Celestial-Order-V2.5, serving as its base.

Merge Details

The model was constructed by merging:

  • Sorihon/CYDR-24B

The merge process utilized specific density and weight parameters for each component model, with a bfloat16 data type for efficiency. The DARE TIES method is known for its ability to effectively combine model weights, aiming to enhance overall performance and capabilities.

Key Characteristics

  • Parameter Count: 24 billion parameters, offering a balance between performance and computational requirements.
  • Context Length: Supports a substantial context window of 32768 tokens, enabling the processing and generation of longer texts while maintaining coherence.
  • Development Method: Leverages mergekit and the DARE TIES technique, indicating a focus on combining and refining existing model architectures rather than training from scratch.

Potential Use Cases

Given its architecture and parameter count, Celestial-Order-24B-2.75 is suitable for a variety of natural language processing tasks, including:

  • Text generation and completion
  • Summarization
  • Question answering
  • General conversational AI applications