nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Tess

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 19, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Tess model is a 27 billion parameter language model developed by nightmedia, created through a NuSLERP merge of several Qwen3.6-27B and Qwen3.5-27B based models, including migtissera/Tess-4-27B. This model integrates diverse capabilities from its contributing components, aiming for broad utility. It is designed for general language tasks, leveraging its merged architecture to provide a balanced performance profile.

Loading preview...

Model Overview

The nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B-F451-Tess is a 27 billion parameter language model developed by nightmedia. It is constructed using a NuSLERP merge method, combining the strengths of multiple Qwen3.6-27B and Qwen3.5-27B base models. Key contributing models include nightmedia/Qwen3.6-27B-Architect-Polaris2-Fable-B, nightmedia/Qwen3.6-27B-Architect-Polaris-Fable-F451, and migtissera/Tess-4-27B, alongside several models from DavidAU and armand0e.

Key Characteristics

  • Merged Architecture: Leverages the NuSLERP merging technique to combine various specialized Qwen-based models, aiming for a comprehensive and balanced performance across different tasks.
  • Performance Metrics: The model's mxfp8 quantization shows a perplexity of 3.797 and a peak memory usage of 34.74 GB, processing approximately 183 tokens/sec. Other quantizations like qx86-hi, qx64-hi, and mxfp4 offer varying trade-offs between perplexity, memory, and token throughput.
  • Component Integration: The merge incorporates models like migtissera/Tess-4-27B, armand0e/Qwen3.6-27B-Fable-5-Experimental, and several instruction-tuned and uncensored variants, suggesting a focus on diverse conversational and creative capabilities.

Use Cases

This model is suitable for a wide range of general-purpose language generation and understanding tasks, benefiting from the combined expertise of its constituent models. Its merged nature implies potential for robust performance in areas such as:

  • Content Generation: Creating diverse textual content.
  • Conversational AI: Engaging in dialogue and instruction following.
  • Reasoning Tasks: As indicated by arc and boolq scores in the provided brainwaves, suggesting capabilities in common sense reasoning and question answering.

Finetunes
4 models