Tether, the stablecoin issuer, is expanding into medical AI. Its QVAC research team released QVAC MedPsy on July 23, a medical language model designed for smartphones and wearables that runs entirely offline, eliminating the need to upload sensitive patient data to the cloud.
Performance Upset: 1.7B Beats Google’s 27B
Benchmarks show the 1.7B parameter version scored an average 62.62 across seven closed medical tests, beating Google’s MedGemma-1.5-4B-it by 11.42 points despite being half its size. In HealthBench Hard, a real clinical benchmark, it outperformed MedGemma 27B, a model 16 times larger. The 4B version scored 70.54, surpassing MedGemma-27B-text.
Tests included USMLE, MedXpertQA, PubMedQA, and AfriMedQA. Performance gains come from staged training: broad medical supervision, high-value clinical reasoning data, and reinforcement learning (RL) on complex cases.
Low Inference Cost, Easy Mobile Deployment
QVAC MedPsy also slashes inference costs. The 4B model generates responses using about 909 tokens (vs. 2,953 for comparable systems), a 3.2x reduction; the 1.7B model cuts compute by 1.7x. Tether released GGUF quantized versions—Q4_K_M format: 1.7B model ~1.2GB, 4B ~2.6GB—suitable for standard mobile hardware.
“We focus on improving efficiency at the model level, not blindly scaling up,” said Tether CEO Paolo Ardoino. “This directly cuts compute, latency, and cost. In healthcare, you can reason where data already exists—inside hospital systems or on devices—without moving sensitive information to the cloud.” He stressed that AI should not be dominated by cloud giants; privacy and local deployment are the future.
QVAC MedPsy charts a path where efficiency, localization, and privacy define performance. Widespread adoption could reshape medical AI infrastructure economics and reinforce Tether’s vision of decentralization and data sovereignty.