Particle.news

Tether Releases VisionPsy-Nano, a Compact Open-Source Vision‑Language Model for Phones

The company says the model is meant to run locally on devices to protect user privacy and reduce latency while enabling new crypto payment use cases.

Overview

  • VisionPsy-Nano, which Tether Data released on July 29, is a roughly 460 million parameter vision-language model that the company designed to run on consumer smartphones.
  • The model ships under an open-source license so developers can inspect, modify, and deploy it without paying licensing fees.
  • Tether’s QVAC SDK includes TurboQuant quantization that the company says can cut memory use up to fivefold and a growing synthetic corpus called QVAC Genesis that expanded from 41 billion to 148 billion tokens for training.
  • Tether published benchmark claims that VisionPsy-Nano leads smartphone vision-language performance and points to earlier MedPsy scores versus Google, but independent testing and peer review are needed to confirm those results.
  • If developers adopt the on-device model and QVAC’s USDT and Bitcoin payment hooks, users could see faster, more private image-and-text features on phones and new micropayment flows on crypto rails, making independent benchmarks and real-world tests the key next milestones.