#El Salvador #San Salvador #VisionPsy-Nano #Tether Data #QVAC

Tether Data Introduces VisionPsy-Nano

Tether Data has officially launched VisionPsy-Nano, an innovative vision-language model that is reshaping the way artificial intelligence operates directly on mobile devices. This end-to-end open-source solution features a compact design of 460 million parameters, specifically optimized for edge deployment, ensuring high performance for everyday consumer smartphones.

Tether's AI research initiative, QVAC, made this announcement on July 30, 2026. Along with its launch, VisionPsy-Nano has set new industry standards with an impressive overall normalized score of 62.3, placing it at the top of its class. This score makes it the highest-ranked model in the under 0.5 billion parameter category, surpassing competitors such as Liquid AI and Hugging Face.

Performance Breakdown

In extensive evaluations, VisionPsy-Nano outperformed established models on 16 of 17 benchmarks, showcasing its adeptness in various task categories. It consistently outclassed models such as Liquid AI's LFM2.5-VL-450M and Hugging Face's SmolVLM2-500M. For developers and researchers, two variants of VisionPsy-Nano are available:

1. VisionPsy-Nano-460M (Best-in-Class Quality): This variant is engineered for reference accuracy across all standard benchmarks at the 460 million parameter scale.
2. VisionPsy-Nano-460M-Flash (Tuned for Latency): This fast-acting version prioritizes ultra-rapid deployment on consumer devices, achieving rapid latency while preserving around 99% of the model's original quality.

Key Features

Despite its smaller size, VisionPsy-Nano excels across four key capabilities:

• - Document Understanding: With advanced OCR functionality, it accurately extracts text from complex formats, including infographics and reports.
• - Visual Perception: The model has the best scene analysis capabilities in its category, providing superior spatial understanding.
• - Reasoning: Class-leading visual reasoning prowess positions VisionPsy-Nano significantly ahead of its peers.
• - Instruction Following: On models like MM-IFEval, it demonstrates stunning reliability, outperforming even bigger models significantly.

Time-Saving Efficiency

The VisionPsy-Nano-460M-Flash variant showcases remarkable speed by generating the first token approximately 19-36 times faster than others in its category, showcasing significant performance improvements across popular devices, such as the Galaxy S25 Ultra and the iPhone 15. This performance enhancement highlights its usability in real-world applications.

Both model versions are released under the Apache 2.0 license, aimed at researchers and educators. They provide three primary inference pathways for immediate utilization:

• - Full-Precision Quality: Available through Hugging Face Transformers.
• - On-Device Mobile Deployment: Allows for seamless use on smartphones with minimal input commands.
• - High-Throughput Server Inference: Enables robust backend operations for intensive applications.

Vision for the Future

Paolo Ardoino, CEO of Tether, expressed the importance of this technology, emphasizing the need for decentralized, powerful AI solutions. He noted the growing concern around centralized data centers and advocated for AI systems that empower individuals rather than large corporations.

"The release of VisionPsy-Nano symbolizes a clear shift towards more localized and efficient AI technology," Ardoino stated. "Our vision involves democratizing AI, enabling users to leverage smart technologies on their existing devices without relying on corporate data center infrastructures."

With this launch, Tether Data is on track to change how consumers interact with AI, ensuring that advanced functionalities are accessible directly on their smartphones, paving the way for a new era of mobile intelligence.

To see VisionPsy-Nano in action across various devices, check out demonstration videos available online.