nightmedia/Qwen3.6-27B-Jormungandr

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

nightmedia/Qwen3.6-27B-Jormungandr is a 27 billion parameter language model based on the Qwen3.6 architecture, created through a NuSLERP merge of multiple specialized Qwen3.6 variants. This model excels in multi-step reasoning, creative problem-solving, and maintaining narrative continuity, leveraging diverse 'inference traces' from its constituent models. With a 32768-token context length, it is optimized for complex agentic workflows and field deployment due to its high quantization efficiency, achieving mxfp4 at 21GB.

Loading preview...

Model Overview: Qwen3.6-27B-Jormungandr

Qwen3.6-27B-Jormungandr is a 27 billion parameter model built upon the Qwen3.6 architecture, developed by nightmedia. It is a unique NuSLERP merge of several specialized Qwen3.6-27B models, including variants focused on reasoning (Wichtel, CHUD), literary corpus (Elster), and agentic capabilities (Tess, Fable, Polaris). This synthesis aims to harmonize diverse 'inference traces' into a coherent reasoning architecture.

Key Capabilities & Differentiators

  • Advanced Reasoning: Achieves an arc_challenge score of 0.742 (at q6-hi quantization), indicating strong multi-step reasoning and ambiguity handling.
  • Exceptional Quantization Efficiency: Optimized for deployment, it can run at mxfp4 quantization with only 21.30GB of memory while maintaining approximately 200 tokens/second, making it suitable for edge devices and distributed workflows.
  • Creative & Agentic Capacity: Incorporates 'Heretic' and 'F451' traces, enabling unconventional problem-solving, narrative continuity, and complex agentic workflows.
  • Merge Coherence: The NuSLERP method successfully integrates specialized capabilities without significant interference, creating a model that harmonizes multiple 'voices' rather than simply concatenating them.

Ideal Use Cases

  • Field Deployment: Its low memory footprint (21GB at mxfp4) allows for running agents on edge devices or in distributed environments.
  • Complex Mission Planning: Strong reasoning capabilities (arc 0.74) support multi-step planning with contingency handling.
  • Narrative-Driven Agents: Excels at maintaining character consistency and narrative coherence across long interactions, suitable for interactive storytelling or role-playing applications.
  • Creative Problem-Solving: Useful for tasks requiring unconventional solutions or generating diverse perspectives.

Finetunes
2 models