Tether, the issuer of the USDT stablecoin, announced the launch of a new version of the USDT stablecoin. framework QVAC Fabric will be released on March 17th this year. This technical tool allows you to train and run artificial intelligence (AI) models with billions of parameters directly on iPhones, Android smartphones, and even computers with consumer graphics cards.
As explained in the launch announcement, development uses Microsoft’s BitNet architecture, which simplifies AI models by reducing numbers to three options: -1, 0, and 1. This process, known as 1-bit quantization, reduces the weight of the file and the power required to process it. Thanks to this, Mobile devices can perform model customization tasks Previously, expensive industrial servers were required.
QVAC fabric acts as the engine that manages these models. This technical background allows the system to work with Vulkan and Metal, technologies that enable the use of a mobile phone’s graphics processing unit (GPU). In tests conducted by the Tether team, On iPhone 16 it was possible to adjust models with up to 13 billion parametersleverage the power of local hardware without relying on the cloud.
This implementation aims to guarantee user privacy, as sensitive data used to train and tune the AI does not leave the device. Being an open source software available on GitHub,Binaries are accessible to all developers Write code to individually integrate these features into your own applications.
Tether’s vision for decentralized AI
Commenting on the launch, Tether CEO Paolo Ardoino said, “Relying on centralized infrastructure to train large-scale language models stagnates innovation, weakens ecosystems, and puts the balance of society at risk. By enabling the meaningful training of large-scale models on consumer hardware, including smartphones, Tether’s QVAC proves that advanced AI can be decentralized.”
This technology reduces RAM usage by up to 90% compared to full precision models. According to published technical data, Fine-tune your model in less than 10 minutes Now available on high-end devices such as the Samsung S25, it sets a precedent in the availability of local AI tools.