Nimbz/Froopert-31B

VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 10, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Nimbz/Froopert-31B is a 31 billion parameter instruction-tuned language model based on the Gemma 4 architecture, developed by Nimbz. This one-step merge model is specifically engineered to reduce the typical 'Gemma-voice' while retaining strong reasoning capabilities and enhancing agentic performance. It excels in complex roleplay scenarios, creative writing with refined word choice, and serves as a robust agent for tasks like crafting image prompts and managing coding workflows.

Loading preview...

Froopert-31B: A Refined Gemma 4 Merge

Froopert-31B is a 31 billion parameter instruction-tuned model, a one-step merge built upon the gemma-4-31B-it base. Developed by Nimbz, the primary goal of this merge was to mitigate the characteristic 'Gemma-voice' while preserving and enhancing its core reasoning abilities and agentic potential. It achieves this by strategically integrating five distinct models, with scotoma-2 acting as a core anchor to manage the base model's stylistic tendencies and improve long-context recall.

Key Capabilities & Differentiators

  • Reduced 'Gemma-voice': Engineered to minimize the base Gemma model's distinctive conversational style, offering a more neutral yet expressive output.
  • Enhanced Reasoning: Maintains and improves upon Gemma 4's inherent reasoning capabilities, making it suitable for complex problem-solving.
  • Agentic Performance: Demonstrates strong performance in agentic workflows, capable of handling tasks like generating image/video prompts, web-search, and code management within a custom harness.
  • Advanced Roleplay: Incorporates Melinoe-VL-heretic for robust in-character roleplay and Dark-Scarlett v2.0 for refined, less cliché word choice.
  • Initiative & Backbone: Integrates Serenity v1.1 to foster greater initiative and reduce excessive agreement, alongside Pantheon-Reasoning 1.1 for improved reasoning in roleplay contexts.
  • Unique Merge Strategy: Utilizes a della_linear merge method with specific layer weighting to ensure the model's intelligence is preserved without 'overcooking' the weights, and a modified 'spine' for long-context handling.

Use Cases

Froopert-31B is particularly well-suited for:

  • Agentic AI Applications: Ideal for developers building agents that require strong reasoning, tool use, and consistent persona adherence.
  • Complex Roleplay & Creative Writing: Excels in generating nuanced, in-character dialogues and creative text with a sophisticated vocabulary.
  • Prompt Engineering: Can effectively craft detailed prompts for various generative AI tasks, including image and video generation.
  • Code Assistance: Capable of assisting with coding tasks, including reading files, executing commands, and making precise edits, while maintaining a distinct persona.

Finetunes
3 models