According to CoinBeat Monitoring, Tether's USDT issuer, the AI research team, announced today the launch of the QVAC MedPsy series of medical language models, designed for localized medical AI on low-power terminals such as smartphones and wearables. It can run without relying on a cloud server, achieving performance far beyond model size through efficient architecture: the 1.7B parameter version averages a score of 62.62 on seven closed medical benchmarks, surpassing Google's MedGemma-4B by 11.42 points, and outperforming the MedGemma-27B with nearly 16 times the parameter size in real clinical scenarios such as HealthBench Hard; the 4B parameter version scores even higher at 70.54, surpassing larger models comprehensively while significantly reducing inference token consumption (up to 3.2 times) and released in quantized GGUF format (1.7B around 1.2GB), suitable for mobile and edge deployment.

This release challenges the traditional assumption of "larger model = better performance," focusing on efficiency through phased medical post-training (supervised, clinical inference data + reinforcement learning) to achieve true local privacy protection and low-latency inference. Tether CEO Paolo Ardoino stated that this allows medical AI to process sensitive data directly on-site at hospitals and device ends without the need to transmit to the cloud, reducing costs, latency, and privacy risks, potentially reshaping the infrastructure of medical AI and promoting local deployment globally, especially in underdeveloped regions.