Tether’s artificial intelligence (AI) research arm has unveiled QVAC Genesis I, the largest synthetic dataset ever created for AI training, comprising 41 billion text tokens.
The dataset is designed to improve reasoning and precision in science and engineering-focused language models, with benchmarks showing strong performance across mathematics, physics, biology, and medicine, according to an emailed announcement on Friday.
Alongside Genesis I, Tether unveiled QVAC Workbench, a local AI app enabling users to run, train, and interact with models directly on their own devices. The app supports leading open models, including Llama, Medgemma, Qwen, and Whisper, while keeping all data private and on-device.
CEO Paolo Ardoino said the twin releases aim to “decentralize intelligence,” moving AI computation from centralized cloud systems to personal hardware. “Intelligence shouldn’t be centralized,” he said. “It should belong to the individual, not the institution.”
Tether has been focusing on decentralized AI for some time, introducing an open-source Wallet Development Kit (WDK) last year to enable humans, machines, and AI agents to build and use secure, self-custodial wallets and transact using USDT and bitcoin .
The stablecoin giant is doing this to diversify their business beyond stablecoins and strategically position itself at the intersection of crypto and decentralized AI infrastructure.