Overview
- QVAC Psy and QVAC MedPsy, released Thursday by Tether’s AI group, are open-source models built to run on phones, wearables, and other low-power hardware, with MedPsy offered in 1.7B and 4B sizes.
- Tether reports that MedPsy-1.7B scored 62.62 and MedPsy-4B scored 70.54 on standard medical tests, beating Google’s MedGemma-4B and even MedGemma-27B on some tasks, though these results have not been independently verified.
- The models ship in quantized GGUF builds sized about 1.2 GB and 2.6 GB and can be downloaded for free on Hugging Face.
- Tether says the 4B model produces answers in about 909 tokens versus roughly 2,953 for comparable systems, which lowers compute needs, speeds responses, and makes on-device use practical for hospitals and mobile users.
- The launch extends Tether beyond stablecoins into privacy-first healthcare AI, yet real clinical deployment would still require external validation and regulatory review given ongoing concerns about unsafe medical advice from large language models.