Tether launches mobile local medical AI: 1.7B small model surpasses 16 times larger models, completely eliminating reliance on the cloud

According to Beating Monitoring, the AI research team of USDT issuer Tether announced today the launch of the QVAC MedPsy series medical language models, a localized medical AI designed specifically for smartphones, wearables, and other low-compute devices. It can run without relying on cloud servers, achieving performance far beyond the model size through an efficient architecture: the 1.7B parameter version scored an average of 62.62 on seven closed medical benchmarks, surpassing Google MedGemma-4B by 11.42 points, and outperformed the parameter-heavy MedGemma-27B (nearly 16 times larger) in real clinical scenarios like HealthBench Hard; the 4B parameter version scored even higher at 70.54, fully surpassing larger models while significantly reducing inference token consumption (up to 3.2 times), and is released in a quantized GGUF format (about 1.2GB for 1.7B), suitable for mobile and edge deployment. This release challenges the traditional assumption that “bigger models = better performance,” focusing on improving efficiency through staged post-training medical fine-tuning (supervised, clinical reasoning data + reinforcement learning), achieving true local privacy protection and low-latency inference. Tether CEO Paolo Ardoino stated that this enables medical AI to process sensitive data directly on hospital premises or device endpoints without transmitting to the cloud, thereby reducing costs, latency, and privacy risks, and has the potential to reshape the infrastructure of medical AI, promoting localized deployment especially in developing regions worldwide.