Sorihon/Celestial-Order-24B-V2.5

TEXT GENERATIONPricing:Input $0.7 / Cached $0.04 / Output $1.16Concurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 25, 2026Architecture:Transformer Featherless Exclusive Cold

Celestial-Order-24B-V2.5 is a 24 billion parameter language model developed by Sorihon, built upon the Celestial-Order-24B-V2 base model. This iteration was created using the DARE TIES merge method, incorporating Delta-Vector/Rei-24B-KTO to enhance its capabilities. With a 32768 token context length, it is designed for general language tasks, leveraging its merged architecture for improved performance.

Loading preview...

Celestial-Order-24B-V2.5 Overview

Celestial-Order-24B-V2.5 is a 24 billion parameter language model developed by Sorihon. This model is a product of a sophisticated merge operation, utilizing the DARE TIES method to combine the strengths of multiple pre-trained models.

Merge Details

The core of Celestial-Order-24B-V2.5 is built upon the Sorihon/Celestial-Order-24B-V2 base model. To further refine its capabilities, it was merged with Delta-Vector/Rei-24B-KTO. This merging process, managed by mergekit, involved specific density and weight parameters for each contributing model, aiming to create a more robust and capable language model.

Key Characteristics

  • Parameter Count: 24 billion parameters, offering a substantial capacity for complex language understanding and generation.
  • Context Length: Supports a context window of 32768 tokens, enabling the processing of longer inputs and maintaining coherence over extended conversations or documents.
  • Architecture: Leverages a merged architecture, specifically the DARE TIES method, which is designed to combine the knowledge and abilities of its constituent models effectively.

Intended Use Cases

This model is suitable for a broad range of general-purpose language tasks, benefiting from its large parameter count and extended context window. Developers can consider it for applications requiring advanced text generation, comprehension, and reasoning, where the combined strengths of its merged components can provide enhanced performance.

Finetunes
1 models