Tether Launches QVAC MedPsy AI for Edge Devices

Tether, the company behind the USDT stablecoin, launched QVAC MedPsy on Thursday, introducing two compact medical AI models designed to run locally on smartphones, wearables and other low-power devices without cloud infrastructure.

The models, developed by Tether's AI Research Group, come in 1.7 billion-parameter and 4 billion-parameter versions. According to Tether, the smaller 1.7B model achieved an average score of 62.62 across seven medical benchmarks, outperforming Google's MedGemma 1.54B by more than 11 points despite being less than half its size. It also surpassed MedGemma 27B, a model nearly 16 times larger, on HealthBench Hard, an OpenAI benchmark evaluating realistic clinical conversations graded by 262 physicians.

The larger 4B model scored 70.54 on the same benchmarks, exceeding MedGemma 27B, which is almost seven times its size. Performance spanned tests including MedQA (USMLE-style questions), MedMCQA, MMLU Health, PubMedQA, AfriMedQA (for African healthcare contexts) and MedXpertQA. Tether attributed the results to a training approach emphasizing efficiency through supervised learning, curated clinical data and reinforcement learning, rather than increasing model...