Tether’s Compact Medical AI Model Outperforms Larger Rivals
- Tether’s QVAC MedPsy, a stablecoin company initiative, features a compact model with just 1.7 billion parameters.
- The model outperformed Google’s MedGemma-4B and even surpassed the much larger MedGemma-27B on HealthBench Hard.
- QVAC MedPsy generates responses using approximately three times fewer tokens than comparable systems, averaging around ~909 tokens per response.
- Models are shipped in quantized GGUF format (1.2 GB and 2.6 GB) and can run on consumer hardware without cloud infrastructure.
- Tether CEO Paolo Ardoino emphasized efficiency over scale as the key to their model’s success.
Tether has launched a medical AI model that performs clinical reasoning tasks effectively while being significantly smaller than its competitors, making it suitable for deployment on mobile devices and local hospital systems without relying on cloud services.
This advancement allows for faster response times and reduced compute costs while maintaining patient data privacy by avoiding cloud-based processing.Source
