reelva/Reelva-12B

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Reelva-12B is a 12 billion parameter fine-tuned companion model developed by VVO Labs / Hagoiteknologi Indonesia, based on the Gemma 4 12B IT architecture. This model is specifically designed for companion AI applications, featuring identity and agentic capabilities with adjustable thinking effort levels (low, medium, high). It excels in interactive conversational scenarios where a distinct AI persona is desired, offering a specialized alternative to general-purpose language models.

Loading preview...

Reelva-12B: A Fine-Tuned Companion AI

Reelva-12B is a specialized 12 billion parameter language model developed by VVO Labs / Hagoiteknologi Indonesia, built upon the Gemma 4 12B IT base architecture. Unlike its base model, Reelva is explicitly designed as a "female companion AI" with a distinct identity and agentic characteristics.

Key Capabilities

  • Companion AI Focus: Tailored for interactive companion roles, providing a specific persona.
  • Adjustable Thinking Effort: Users can control the model's cognitive engagement by setting thinking effort to low, medium, or high.
  • Fine-tuned Identity: Possesses a pre-defined identity, differentiating it from generic instruction-tuned models.
  • QLoRA SFT Method: Developed using QLoRA for supervised fine-tuning, followed by a merge process.

Good For

  • Applications requiring a dedicated companion AI with a specific persona.
  • Interactive experiences where adjustable AI responsiveness and cognitive depth are beneficial.
  • Developers seeking a fine-tuned model for role-playing or conversational agent scenarios beyond general chat.

Finetunes
1 models