Tether’s QVAC Genesis I Redefines AI Training with the World’s Largest Synthetic STEM Dataset

In a decisive move toward democratizing artificial intelligence, Tether Data’s research arm, QVAC, has launchedQVAC Genesis I—the world’s largest and most advanced synthetic dataset designed to train STEM-focused AI models. Comprising an extraordinary 41 billion text tokens, this release marks a defining moment for open, community-driven AI development outside the confines of Big Tech.

Each token within the Genesis I dataset represents a small fragment of language—the atomic unit from which AI models learn to interpret and generate text. By analyzing billions of these building blocks, the dataset allows models to develop not only linguistic fluency but also logical and scientific reasoning across mathematics, physics, biology, and medicine. Early validation shows that Genesis I-trained systems exhibit significantly improved problem-solving accuracy and conceptual understanding in STEM disciplines where existing public datasets have long fallen short.

More than a technological milestone, Genesis I is an ideological statement. “Intelligence shouldn’t be centralized,” said Paolo Ardoino, CEO of Tether. “With QVAC Workbench and Genesis I, we’re opening the door to infinite intelligence—AI that lives, learns, and evolves locally on your own device.” His vision challenges the current AI ecosystem dominated by massive data monopolies and proprietary models.

To complement the dataset, Tether has also unveiledQVAC Workbench, a robust application that enables users to run large language models directly on their devices. The app currently supports leading open models such as Llama, Qwen, MedGemma, Whisper, and SmolVLM, and is available on Android, with iOS, Windows, macOS, and Linux releases imminent. Unlike cloud-based AI services, all user interactions remain on-device, ensuring complete data ownership and privacy.

A distinctive feature, “Delegated Inference,” allows users to link their mobile and desktop versions of Workbench, pooling computational resources through a peer-to-peer network. This design echoes Tether’s broader goal: decentralizing computation and restoring agency to the individual user.

The QVAC Genesis initiative is built on a rigorous multi-stage data generation process that transforms verified scientific and educational materials into structured learning resources. This approach aims to cultivate models capable of critical reasoning rather than surface-level text imitation. “Most AI today sounds smart but doesn’t truly think,” Ardoino noted. “We designed Genesis to help models understand cause and effect and reason their way through complexity—and we’re making it open to everyone.”

Tether Data’s move represents a bold reimagining of AI’s future—one grounded in openness, scientific rigor, and individual empowerment. With Genesis I and Workbench, the company positions itself not as a competitor to corporate AI labs but as a catalyst for a new paradigm:local intelligencethat is free, private, and globally accessible.