Lythri/Lythri-4B-A2B

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:5.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Lythri/Lythri-4B-A2B is a 5.1 billion parameter on-device language model developed by Lythri, built on Gemma 4 E2B with a 32768 token context length. It is specifically optimized for emotional companionship, understanding human feelings, and engaging in natural, multi-turn conversations. This model is designed to run locally on personal devices, prioritizing emotional intelligence over traditional math or coding benchmarks. It achieves emotional understanding comparable to larger models while maintaining a small footprint.

Loading preview...

Lythri-4B-A2B: On-Device Emotional Companionship Model

Lythri-4B-A2B is a 5.1 billion parameter language model (2.3B active parameters) from the Lythri family, built upon the Gemma 4 E2B architecture. Unlike many LLMs focused on academic benchmarks like math or coding, Lythri is specifically designed for emotional companionship and understanding human feelings, enabling natural, multi-turn conversations. Its small size allows it to run efficiently on local devices such as laptops or phones.

Key Capabilities & Differentiators

  • Emotional Intelligence: Lythri-4B-A2B demonstrates strong performance on emotion benchmarks, achieving results comparable to the larger Gemma 4 E4B instruction-tuned model (8B parameters) despite its significantly smaller active parameter count.
  • On-Device Operation: Optimized for local deployment, making it suitable for applications requiring privacy or offline functionality.
  • Conversational Fluency: Trained to engage in natural, multi-turn dialogues with a focus on understanding and responding to emotional nuances.
  • Balanced Performance: While excelling in emotional understanding, it maintains reasonable performance across general benchmarks for knowledge and reasoning, though it is not optimized for complex math or coding tasks.

Ideal Use Cases

  • Personalized AI Companions: Developing applications that require empathetic and emotionally aware conversational agents.
  • Mental Wellness Support: Providing non-professional, AI-driven emotional support and conversational interaction.
  • Interactive Storytelling & Roleplay: Creating characters that can respond with emotional depth and understanding.
  • Edge AI Applications: Deploying conversational AI on devices with limited computational resources.

Finetunes
4 models