Wu said that Tether's AI research team launched the QVAC MedPsy medical language model today. The series of models are designed to run directly on low-power devices such as smartphones and wearable devices, providing performance comparable to large models while achieving full localization and privacy protection. Data shows that its 1.7 billion parameter model scored an average of 62.62 across seven medical benchmark tests, outperforming Google MedGemma-1.5-4B-it, which has more than twice the number of parameters, by 11.42 points. Tether CEO Paolo Ardoino stated that this move aims to change the limitations of medical AI applications, enabling medical reasoning to run locally where the data resides (such as hospital systems or mobile devices), without relying on cloud processing of sensitive information.