'Intelligence shouldn't be centralized': Tether's AI arm releases open dataset and local app

Tether Data's research division, QVAC, has launched a new initiative called Genesis, describing it as the world's largest synthetic dataset for artificial intelligence training, alongside Workbench, a local AI app that enables private on-device model use, in a push to decentralize artificial intelligence.

In a statement shared with The Block, Tether said AI is becoming increasingly centralized, trained, hosted, and controlled by a handful of corporations. The initiative seeks to level the playing field, enabling open, community-driven intelligence to "thrive outside Big Tech's walls."

The first release, Genesis I, includes 41 billion tokens — the smallest units of text that AI models learn from — to help build smarter, more capable, and more precise STEM-focused language models. Training on this data helps LLMs better understand words, logic, and problem-solving in science, technology, engineering, and math, the firm said.

Meanwhile, Tether Data's first consumer app, QVAC Workbench, is a local AI app for enthusiasts, advanced users, and researchers that runs models directly on their own devices, rather than relying on Big Tech cloud infrastructure. It supports many leading open models, including Llama, Medgemma, Qwen, SmolVLM, and Whisper, and is available on Android, Windows, macOS, and Linux, with iOS to follow.

Helping AI think critically rather than merely imitate language

"Intelligence shouldn't be centralized," Tether CEO Paolo Ardoino said. "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. We believe that intelligence, like information, should be free, accessible, and owned by everyone, not locked behind corporate firewalls or sold as a service. Whether it's a phone, a robot, or a wearable, intelligence should belong to the individual, not the institution."

Tether Data argues that by making the QVAC Genesis dataset public, researchers can build models that rival or even surpass proprietary systems, offering a training resource that helps AI reason, solve problems, and think critically rather than merely imitate language.

"Most AI today sounds smart, but doesn't truly think," Ardoino continued. "We designed this dataset to help models understand cause and effect, to make connections, draw conclusions, and reason their way through complexity. And we're making it open to everyone."

Tether, whose team of less than 200 generated a profit of approximately $13 billion in 2024 and currently holds more than 100,000 BTC ($10 billion), has been aggressively deploying capital into a variety of projects beyond stablecoins in recent years across sectors, including AI, as well as telecommunications, data centers, energy infrastructure, DATs, and bitcoin mining.