Tether Launches Local Medical Large Model QVAC MedPsy

The Tether AI research team has released the medical language model QVAC MedPsy, which can run locally on low-computation devices such as smartphones and wearables without relying on the cloud. According to the official statement, the 1.7B parameter model achieved an average score of 62.62 across seven closed medical benchmark tests, outperforming Google MedGemma-1.5-4B-it by 11.42 points. Additionally, it surpassed MedGemma 27B in clinical scenarios such as HealthBench Hard. The 4B version scored 70.54 in the same group of tests, outperforming competitors approximately seven times larger in scale, while reducing the average number of generated tokens to about one-third to one-half compared to comparable systems, thereby lowering inference costs and latency.