Tether Data has released QVAC MedPsy, a specialized medically-oriented artificial intelligence (AI) model built on its foundational QVAC Psy framework, which is based on the premise of “Psychohistory.” The models are designed for sovereign, private medical intelligence that runs entirely on local hardware, no cloud required.
How QVAC MedPsych is made to work
The QVAC MedPsyc family of models consists of two types of models: a fully-precise BF16 model [(BFloat16) model that uses 16-bit Brain Floating Point] and a (GPT-Generated Unified Format (GGUF)-quantized model for local inference. Across seven medical benchmarks, the 1.7B model outperforms the well-known Google’s MedGemma 4B by over 11 points and even beats the sixteen-times-larger MedGemma 27B on HealthBench Hard. The 4B model outperforms the nearly seven-times-larger MedGemma 27B on the same benchmarks. This “superior methodology,” according to Tether, reverses the parameter gap.
QVAC MedPsy’s main features
- Completely open-source: Weights available for full audit and adaptation
- Local execution: Runs on consumer Graphics Processing Unit (GPUs) using QVAC Fabric
- Massive token efficiency: Uses 3.2x fewer tokens than backbone models
- Quantized variants: 4-bit version preserves accuracy within around 1 point while reducing footprint by around 69 percent
Privacy matters and use cases
Because QVAC MedPsy runs fully on local hardware, sensitive patient data/records are stored within the device. This makes it ideal for:
- Clinical decision-making for hospital patients with rigid data sovereignty requirements
- Diagnostic assistance in low bandwidth or remote settings/locations
- Personal health advisors who rely/run fully on smartphones or laptops
- Medical education and practice/simulation on consumer devices
According to Tether’s vision as articulated in the announcement, “8 billion humans deserve an intelligence that doesn’t blink when the signal dies,” meaning AI that works offline, private, and uncoupled from a centralized system.
Transforming healthcare
There has not been much rapid progress in AI implementation within the healthcare sector, even with the most advanced AI tech developments nowadays. This slow adoption is primarily due to privacy concerns, latency issues, and dependency on cloud infrastructure. QVAC MedPsy addresses all three aspects: it runs locally (no data leaves the device), achieves near-instant inference [critical for emergency settings like an emergency room (ER)], and requires no internet connection. This will change the way healthcare is delivered in areas with limited resources, such as rural clinics, field hospitals, and developing countries.
