Tether Launches Compact Edge AI for Medical Reasoning Models

• Tether launches compact MedPsy AI models that run locally, reducing reliance on cloud systems.
• New MedPsy models show strong benchmark results while focusing on privacy and faster medical reasoning.
• Edge-based AI design aims to cut costs, improve speed, and support hospitals handling sensitive data.

Tether has released new medical AI models that can run directly on phones, laptops, and local servers without relying on cloud systems. The company’s AI research team introduced QVAC MedPsy in 1.7 billion and 4 billion parameter versions, designed to handle medical reasoning tasks while keeping all data on the device. 

In an official release, the firm said that the models mainly target hospitals, clinics, and developers that need faster processing and tighter control over patient information.

Tether said the models compete with larger cloud-based systems in medical tests despite their smaller size. It also stated that the 1.7B version outperforms several mid-sized competitors in clinical benchmarks. In addition, the 4B model records stronger results than some much larger systems in selected medical evaluations.

Edge AI Breakthrough in Medical Reasoning

Tether designed the MedPsy models to run directly on edge devices without relying on cloud infrastructure. The company said this setup reduces delays and improves privacy when handling healthcare data. The 1.7B model also outperformed Google’s MedGemma-1.5-4B-it by 11.42 points in benchmark testing. In addition, it showed strong results in HealthBench and other clinical reasoning evaluations.

The 4B version also delivered higher scores than MedGemma-27B-text-it across several medical tests. It recorded improvements in MedXpertQA and HealthBench Hard benchmarks as well. Tether said it trained the models using reinforcement learning combined with curated clinical data. This approach, according to the company, improves accuracy in complex medical reasoning tasks.

Benchmark Gains Over Larger Models

QVAC MedPsy delivered strong results across several medical benchmarks, including MedQA, MedMCQA, and PubMedQA. It also performed well on AfriMedQA and MMLU health datasets, which test medical knowledge across different regions and use cases.

Tether said the system uses fewer tokens per response, which helps reduce computing costs and improve speed. The 1.7B model averages about 1,110 tokens per response, while the 4B version averages around 909 tokens in tests. As a result, the company says the models can run efficiently on standard hospital hardware. It also provides compressed versions for mobile and edge device deployment.

Impact on Healthcare Privacy and Costs

Tether said running AI models locally can reduce privacy risks linked to cloud-based medical systems. It added that on-site processing helps hospitals manage sensitive patient data without sending it to external servers.

Separately, the company recently reported strong financial performance and stable circulation of its USDT stablecoin. It also said it continues to work with regulators, including actions such as wallet freezes and preparations for audits.

Related: UAE Builds Regulated Rails for Swapping Dirham and Dollar Stablecoins