KordAI/Qwen-Sama-4B

VISIONConcurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 5, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

KordAI/Qwen-Sama-4B is a 4.5 billion parameter conversational AI model, fine-tuned from Qwen/Qwen3.5-4B, specifically optimized for VTuber, livestream, and virtual companion roleplay. It excels at generating engaging, natural, and context-aware responses for interactive dialogue applications. The model maintains the strong reasoning and language capabilities of its Qwen 3.5 base while being lightweight enough for local inference.

Loading preview...

Qwen Sama 4B: Conversational AI for Virtual Companions

Qwen Sama 4B, developed by KordAI, is a 4.5 billion parameter conversational AI model built upon the robust Qwen/Qwen3.5-4B architecture. It is specifically instruction-tuned using Supervised Fine-Tuning (SFT) on synthetic conversational datasets to deliver highly engaging and context-aware interactions.

Key Capabilities

  • Virtual Streamer Personality: Designed to embody a virtual streamer or companion persona.
  • Natural Multi-Turn Conversations: Generates fluid and coherent dialogue over extended interactions.
  • Roleplay-Optimized Dialogue: Excels in character roleplay scenarios, producing appropriate and immersive responses.
  • Context-Awareness: Utilizes conversation history to maintain relevance and consistency.
  • Lightweight Inference: Built on a 4B parameter base, making it suitable for local deployment.

Intended Use Cases

This model is optimized for applications requiring interactive and personalized conversational AI, including:

  • VTuber chatbots and AI streamers
  • Virtual companion applications
  • Interactive Discord bots and streaming assistants
  • Character roleplay scenarios

For optimal performance, users should provide a system prompt that clearly defines the desired character or personality. While strong in conversational roleplay, the model is not optimized for factual question answering or complex technical tasks, and responses may occasionally contain hallucinations. Human moderation is recommended for public-facing deployments.