Tether QVAC releases 460M smartphone vision model
QVAC, Tether Data’s AI research unit, open-sourced VisionPsy-Nano, a 460-million-parameter vision-language model that runs on phones and analyzes images and documents offline.
On July 29, 2026, QVAC, the AI research arm of Tether Data, published VisionPsy-Nano, a 460-million-parameter vision-language model designed to run on smartphones and perform image and document analysis without sending data to remote servers.
The model processes photos, scanned documents, flowcharts and infographics. It extracts text with optical character recognition and answers user queries on-device. QVAC released two builds: a standard version that prioritizes accuracy and a Flash version that reduces a small amount of quality to produce much faster initial responses on mobile hardware.
In benchmark testing across 17 vision-language tasks, VisionPsy-Nano led on 16. The model posted a normalized score of 62.3. By comparison, Liquid AI’s LFM2.5-VL-450M scored 59.6 and SmolVLM2-500M scored 52.5; QVAC’s earlier nanoVLM-460M-8k scored 54.9. QVAC reported the new model outperformed similarly sized competitors by an average of 7.4% on visual reasoning tests and exceeded models more than twice its size on MM-IFEval and POPE, including releases from Qwen and InternVL.
QVAC highlighted the Flash build’s latency improvements as a practical advantage on phones. Flash retains about 99% of the full model’s performance while producing its first output up to 23 times faster than SmolVLM2-500M on Android devices and up to 36 times faster on an iPhone 15, according to QVAC. The company cited device limits such as memory, heat and battery as reasons for optimizing initial response time.
Developers can load VisionPsy-Nano through Hugging Face Transformers, a quantized GGUF format compatible with llama.cpp, or a vLLM backend intended for higher-volume server use. QVAC published its benchmark configurations via VLMEvalKit to allow independent reproduction of the tests. Both builds are released under the Apache 2.0 license, which permits commercial use, modification and redistribution.
In a statement, Tether CEO Paolo Ardoino wrote, “Achieving best-in-class quality and performance on general vision tasks at just 460 million parameters proves that local-first, highly efficient AI is a viable pathway.” The model follows QVAC releases dating to October 2025, including synthetic training datasets, a cross-platform software development kit and medical models for phones and wearables.
Tether has funded its AI expansion with revenue from its USDT stablecoin business and investments in AI infrastructure, biotechnology and robotics, including a Series C investment in NEURA Robotics valued at up to $1.4 billion. QVAC said VisionPsy-Nano provides a way for developers and businesses to run offline image analysis without per-use API fees and for users to keep sensitive documents on-device while the phone handles both input and output.
The material on GNcrypto is intended solely for informational use and must not be regarded as financial advice. We make every effort to keep the content accurate and current, but we cannot warrant its precision, completeness, or reliability. GNcrypto does not take responsibility for any mistakes, omissions, or financial losses resulting from reliance on this information. Any actions you take based on this content are done at your own risk. Always conduct independent research and seek guidance from a qualified specialist. For further details, please review our Terms, Privacy Policy and Disclaimers.
Articles by this author
