Tether unveils QVAC MedPsy: compact medical AI models

Tether launched QVAC MedPsy, compact medical AI models for edge devices, outperforming larger rivals on benchmarks while running locally and preserving user privacy.

Tether has introduced QVAC MedPsy, a new suite of medical AI language models optimized for smartphones, wearables, and other edge devices. Unlike traditional models that rely on cloud infrastructure, QVAC MedPsy operates locally, ensuring enhanced privacy and efficiency. The models come in 1.7 billion and 4 billion parameter versions, both delivering expert-level healthcare reasoning. Notably, they outperform much larger state-of-the-art models like Google’s MedGemma-4B and MedGemma-27B on key medical benchmarks. The 1.7B model scored 62.62 across seven closed-ended benchmarks, surpassing MedGemma-4B by over 11 points, while the 4B model achieved 70.54, outperforming models nearly seven times larger. QVAC MedPsy models generate responses using up to 3.2 times fewer tokens than comparable systems, enabling faster and more efficient operation on consumer hardware. Distributed in quantized GGUF format, the models are compact (1.2 GB and 2.6 GB), privacy-focused, and suitable for clinical, low-resource, and personal health applications. Released under the Apache 2.0 license, they are available on Hugging Face.