Tether Data open-sources VisionPsy-Nano vision-language model with 460 million parameters

Tether Data, the technology arm of the company best known for issuing the world’s largest stablecoin, just dropped an open-source AI model designed to run on your phone. VisionPsy-Nano, a vision-language model packing roughly 460 million parameters, was released on July 29 as part of Tether’s QVAC AI research initiative. The company claims the model leads industry benchmarks for consumer smartphone AI. From stablecoins to neural networks Since March 2024, Tether Data has been building toward a future where AI models run locally on user devices rather than bouncing queries off massive cloud servers. The logic is straightforward: local execution means better privacy, lower latency, and no dependence on centralized infrastructure. VisionPsy-Nano is a “vision-language model,” which means it processes both images and text. At 460 million parameters, it’s deliberately compact. For comparison, the largest frontier models from major AI labs run into the hundreds of billions or even trillions of parameters. The model ships under an open-source license, meaning developers can inspect, modify, and deploy it without licensing fees. The QVAC pipeline and synthetic data play The QVAC Genesis ser...