Key Takeaways:
Tether has unveiled two new open-source AI models – QVAC Psy and QVAC MedPsy–specifically created for devices like phones and edge devices.According to the company, the 1.7B and 4B product models beat out a few significantly bigger medical artificial intelligence tools.Private, local processing of Artificial Intelligence without the need of cloud infrastructure is the key theme in QVAC MedPsy.
Tether is expanding beyond crypto and stablecoins with a new push into artificial intelligence. The company also unveiled an AI initiative called QVAC Psy, which is a basic AI product, and QVAC MedPsy, an AI language model for the healthcare industry intended for deployment directly on smartphones, wearables and low power devices.
The launch underscores Tether’s evolving trajectory towards investing in local-first AI systems and decentralized pipelines.
8 billion humans deserve an intelligence that doesn’t blink when the signal dies. 🧠Introducing @QVAC Psy, our foundational models built on the mathematical stability of Psychohistory.
With QVAC MedPsy, our local-first medical health AI model, we’ve proven that superior… pic.twitter.com/6ECt7kvk6Q
— Tether (@tether) May 7, 2026
Read More: Tether Rolls Out New Self-Custodial Wallet
Tether Launches Local-First Medical AI
According to Tether, QVAC MedPsy was built to reduce dependence on centralized cloud servers. Instead of requiring hospitals or users to send sensitive medical data through remote infrastructure, the models can process information directly on-device. The company released two versions:
QVAC MedPsy 1.7B QVAC MedPsy 4B
Both are optimized for local deployment and lower computing requirements. Tether said the models are fully open-source and available in compressed formats suitable for consumer hardware.
Smaller Models Deliver Strong Results
Tether claims its compact AI systems achieved benchmark scores competitive with much larger models from major tech firms.
Efficiency Over Raw Size
The 1.7B model reportedly scored 62.62 across seven medical benchmark suites, outperforming Google’s MedGemma-1.5-4B-it despite being less than half the size. Meanwhile, the 4B version achieved a 70.54 average score across the same tests and reportedly exceeded some models nearly seven times larger in clinical-style evaluations.
The benchmark testing included:
MedQA-USMLE MedMCQA PubMedQA HealthBench MedXpertQA
Tether said its training approach focused on medical reasoning efficiency instead of simply increasing parameter count.
Read More: Tether Selects KPMG for First Full USDT Audit
Lower Costs and Faster Processing
The company also highlighted reduced inference costs and faster response generation. According to Tether, the 4B model generated responses using roughly 909 tokens on average, compared to more than 2,900 tokens for comparable systems. This results in reduced compute costs and increased output as fewer tokens are used.
This allows the models to run locally, reduces reliance on the internet and expensive cloud infrastructure, while also keeping heavy workloads off the cloud.
Tether Expands Beyond Stablecoins
The launch represents a new policy unprecedented for Tether. The company also has an AI arm, Tether Data, developing AI infrastructure that is decentralized and private, in addition to its crypto venture.
Locally-developed AI could transform the healthcare landscape by eliminating the need for hospitals and institutions to transfer critical patient data outside safe settings, said chief executive Paolo Ardoino.
Tether positioned QVAC Psy as part of a future where AI systems operate directly on personal devices with greater privacy, lower costs, and reduced reliance on centralized tech platforms.
