Tether Data, the artificial intelligence division of the stablecoin issuer, has open-sourced VisionPsy-Nano, a 460-million-parameter vision-language model. The company says the release could reshape how mobile devices handle AI tasks by keeping data on-device rather than shipping it to the cloud — a pitch that resonates in both the crypto and broader tech industries.
What the model does
VisionPsy-Nano is designed to process images and text together, enabling tasks like visual question answering and image captioning directly on a smartphone or edge device. At 460 million parameters, it's small enough to run locally without a constant internet connection, which Tether Data argues improves both privacy and efficiency. The model is now available on GitHub under an open-source license.
Tether's move into AI has been building for months. The company has been hiring researchers and positioning itself as a player in decentralized AI — a space where crypto-native firms argue that open models can compete with closed systems from Big Tech. By open-sourcing VisionPsy-Nano, Tether Data is putting its code in the public domain, letting developers inspect, modify, and deploy it without permission. That's a direct contrast to proprietary models like OpenAI's GPT-4 or Google's Gemini.
The model is out now, but adoption will depend on how well it performs against other small models like Microsoft's Phi-3 or Meta's Llama 3.2. Tether Data hasn't announced any specific partnerships yet, but the open-source release is a clear signal that the company wants developers to build on its stack. Whether that translates into real-world use — or just another repo on GitHub — remains to be seen.
