Tether pushes compact on-device AI models that run without a signal

Tether is advancing a family of compact models up to ~4 billion parameters (MedPsy, TranslatePsy, VisionPsy) that run locally on smartphones without an internet connection. The company relies on quantization and its specialized Genesis III dataset of 191 billion tokens, while its QVAC SDK cuts KV-cache size by up to 80%.

• Tether's models up to 4B parameters run offline on a basic smartphone
• Genesis III dataset spans 191 billion tokens across 19 STEM domains
• QVAC SDK shrinks model KV cache by up to 80% via TurboQuant
• SDK adds clustered inference to pool multiple machines