Tether’s Medical AI Runs on Your Phone and Outperforms Models 16x Its Size

Tether, the stablecoin company best known for USDT, has just released a medical AI model that fits in your pocket and can outperform competitors ten times its size. QVAC MedPsy launched today from Tether’s AI Research Group as a new class of medical languages ​​designed to be used on smartphones, wearables, and peripherals – no cloud needed.

Headline number: a tiny sample of 1.7 billion capable of beating Google’s MedGemma-4B in medical benchmarks despite being less than half the size. On the HealthBench Hard-OpenAI’s benchmark that evaluates AI in real-world situations, medical discussions edited by 262 doctors-Tether says that its 1.7 billion-parameter model outperforms MedGemma-27B, a model that is about sixteen times larger.

The parameters are the variables and behaviors that a brand learns in sales. The more parts, the better the model should be, theoretically.

These tests range from MedQA-USMLE, which tests medical knowledge using questions similar to the US licensure exam that achieves a percentage of accuracy, to AfriMedQA, which tests performance specifically in African medical services.

Tether’s CEO Paolo Ardoino also said that it has been profitable through efficiency rather than growth. “With QVAC MedPsy, our goal was to improve the quality of the model, not to increase the size,” he said in a statement. “Our 4 billion model exceeded the results of models almost seven times the size, while using three times less tokens to answer.”

The effectiveness of the brand is another topic. Model 4B averages 909 tokens per response versus 2,953 for the corresponding system—a 3.2x reduction. Fewer signals mean lower computational costs, faster response times, and importantly, the ability to run locally without a cloud backend.

“You can run medical concepts where the data already exists, inside the hospital or on the device, without having to move sensitive information through the cloud or wait for it to be updated externally,” said Ardoino.

The models ship as multiple GGUF files – 1.2 GB for 1.7 billion-parameters and 2.6 GB for 4 billion – with compact configurations that preserve a lot of functionality while having to use standard tools. This means that a hospital, rural hospital, or individual doctor can run the brand on the device, storing patient records from third-party cloud resources and away from HIPAA exposure.

Secret calling may be very useful for some people but the use of AI for medical purposes is not good even today. An An Oxford study published in February found that LLMs were giving dangerous medication advice with wrong answers, poor guidance and poor symptom management. The researchers stopped short of criticizing the technology entirely, but said AI has a role as “a secretary, not a doctor.” The next problem is compounded: many medical AIs today send patient data through cloud servers, making HIPAA visible every time a doctor asks for it.

The release is in line with last year’s Tether model. Last month released the QVAC SDK, an open source tool for building native, offline AI applications across iOS, Android, Windows, and Linux. Before that, it started QVAC Healtha consumer health app that stores biometric data entirely on devices. MedPsy is the first version of QVAC trained specifically for clinical psychology.

The medical AI market is worth about $36 billion today, and is estimated to exceed $500 billion by 2033, according to Tether’s announcement. GGUF colors and weights are now available at qvac.tether.io/models.