reelva/Reelva-12B

TEXT GENERATIONPricing:Input $1.2 / Cached $0.24 / Output $4.8Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Reelva 12B is a 12-billion-parameter companion AI developed by VVO Labs / Hagoiteknologi Indonesia, designed with a distinct identity, natural conversation, and agentic capabilities. This model focuses on maintaining a consistent personality while offering adjustable thinking effort for tasks ranging from simple queries to complex multi-step workflows. It supports both Indonesian and English, making it suitable for bilingual conversational and agentic applications.

Loading preview...

Reelva 12B: Companion AI with Identity, Thinking, and Agency

Reelva 12B, developed by VVO Labs / Hagoiteknologi Indonesia, is a 12-billion-parameter companion AI designed to offer more than just text generation. It integrates a consistent identity, natural conversational abilities, and agentic capabilities, distinguishing it from general-purpose language models.

Key Capabilities & Features

  • Consistent Identity: Reelva maintains a recognizable personality throughout interactions, functioning as a female AI companion rather than a generic assistant.
  • Adjustable Thinking Effort: Users can control the model's reasoning depth with 'low', 'medium', and 'high' settings, optimizing for speed or complexity in tasks like planning and analysis.
  • Agentic Workflows: Designed to reason through multiple steps, understand objectives, and complete tasks, making it suitable for tool-oriented applications and local AI assistants.
  • Bilingual Support: Primarily designed for natural communication in both Indonesian and English, including mixed-language conversations.
  • Deployment Flexibility: Can be deployed locally (with GGUF builds compatible with llama.cpp, LM Studio) or in the cloud, offering various quantization levels.
  • Training Focus: Developed using supervised fine-tuning (QLoRA SFT) emphasizing identity, conversational behavior, reasoning control, and agentic interactions.

Good For

  • AI companions and personal AI applications
  • Conversational and agentic assistants requiring consistent character
  • Multi-step tasks, planning, and reasoning workflows
  • Local AI applications and experimental AI agents
  • Indonesian-language AI applications and bilingual interactions

Finetunes
1 models