sophosympatheia/Glistening-Gem-31B-v2.1

VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 16, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Glistening-Gem-31B-v2.1 by sophosympatheia is a 31 billion parameter language model, merged from Artemis-31B-v1, G4-MeroMero-v2-31B, and gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic, using google/gemma-4-31B-it as a base. This model is specifically optimized for creative writing and prose generation, improving upon previous versions in output quality. It is designed for tasks requiring imaginative and expressive text, with a context length of 32768 tokens.

Loading preview...

Overview

sophosympatheia/Glistening-Gem-31B-v2.1 is a 31 billion parameter language model created through a creative merge of several pre-trained models, including TheDrummer/Artemis-31B-v1, zerofata/G4-MeroMero-v2-31B, and llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic, built upon a stock google/gemma-4-31B-it base. This v2.1 iteration specifically enhances creativity and prose generation, addressing issues found in its predecessor, v2.0.

Key Capabilities

  • Enhanced Creativity and Prose: The model is fine-tuned to produce more imaginative and expressive text, making it suitable for creative writing tasks.
  • Improved Stability: Version 2.1 mitigates artifacting issues (like fused or misspelled words) present in earlier versions by using a stable base model and strategic layer management, though rare occurrences may still appear.
  • Optimized for Creative Tasks: The merge ingredients are selected to maximize creative output, making it less suitable for tasks requiring strict logical coherence or "thinking" modes.

Known Limitations

  • Occasional Artifacts: Rarely, the model may produce minor text artifacts such as fused words or misspellings. These can be reduced with conservative sampler settings.
  • Not for Logical Reasoning: The model is not recommended for tasks requiring strong logical thinking or coherence, as its focus on creativity tends to negatively impact such capabilities.

Sampler Tips

Recommended sampler settings are provided as a starting point for creative tasks, including specific values for Temperature, Min-P, Adaptive-P, and DRY parameters. These settings are designed to optimize the model's creative outputs.

Finetunes
2 models