Tether Data Open-Sources VisionPsy-Nano: Best-in-Class ~460M On-Device Vision-Language Model Leading Industry Benchmarks

Tether AI Research launches open-weight, edge-optimized vision-language models, securing the No. 1 spot for quality and performance on everyday consumer smartphones

29 July 2026 – Tether Data’s AI research initiative, QVAC, today announced the open-source release of VisionPsy-Nano, a compact 460-million-parameter vision-language model (VLM) purpose-built for on-device and edge deployment. Bringing multimodal understanding that previously required cloud infrastructure directly onto mobile devices, VisionPsy-Nano establishes a new state-of-the-art for compact VLMs, achieving the highest overall normalized score (62.3) and the number 1 spot for quality and performance among all evaluated sub 0.5B on-device vision-language models.

VisionPsy-Nano consistently ranks as the best-performing model in the ~0.5B VLM category, outperforming competing compact models from Liquid AI and Hugging Face. Across testing, VisionPsy-Nano-460M compared favorably to every other 0.5B VLM tested – including Liquid AI’s LFM2.5-VL-450M, Hugging Face’s SmolVLM2-500M, and its own base model (nanoVLM-460M-8k) – on 16 of 17 benchmarks, achieving an industry-leading overall normalized score of 62.3 (versus 59.6, 52.5, and 54.9, respectively).

To meet different deployment needs, Tether is releasing the model in two distinct variants:

• VisionPsy-Nano-460M (Best-in-Class Quality): Optimized to deliver class-leading, reference accuracy across all standard vision-language benchmarks at the ~460M scale.
• VisionPsy-Nano-460M-Flash (Tuned for Latency): Specifically engineered for ultra-fast real-world deployment on everyday smartphones, delivering massive latency gains while retaining ~99% of full model quality (61.4 normalized score).

Core Capabilities of VisionPsy-Nano

Despite its compact size, VisionPsy-Nano tops all four evaluation capability categories, excelling at complex, multimodal tasks directly on-device:

• Document Understanding & OCR: Extracts text and structural insights accurately from complex backgrounds, flowcharts, financial reports, and infographics.
• Visual Perception: Superior scene analysis and spatial layout understanding, leading the category with a +4.6% relative margin over the next-best ~0.5B model.
• Reasoning & Knowledge: Class-leading visual reasoning that outpaces competing ~0.5B models by a +7.4% relative margin.
• Instruction Following & Reliability: Punches above its weight – on MM-IFEval (42.3) and POPE (87.9), VisionPsy-Nano-460M beats models 1.6x to 2.3x its size, including FastVLM-0.5B (759M), Qwen3.5-0.8B (873M), and InternVL3.5-1B (1061M).

Ultra-Fast Edge Speed & Developer Accessibility

VisionPsy-Nano-460M-Flash pairs best-in-class quality with the fastest on-device deployment in its class. By using far fewer visual tokens, VisionPsy-Nano-460M-Flash reaches first-token generation ~19–23x faster than SmolVLM2-500M and its own base model on Pixel 9, Galaxy S23, and Galaxy S25 Ultra, and up to ~36x faster on iPhone 15. Beyond outperforming the SmolVLM2-500M and our own base model, this responsiveness advantage persists across all four devices, securing lower latency than both the LFM2.5-VL-450M and Qwen3.5-0.8B.

Both variants ship open-weight under the Apache 2.0 license and are intended to support researchers and for educational purposes. There are three inference paths available on day one:

• Full-Precision Quality: Accessible via Hugging Face Transformers.
• On-Device Mobile Deployment: Quantized GGUF checkpoints designed to run on phones through `llama.cpp` using a single-line invocation.
• High-Throughput Server Inference: Production-ready backend inference powered by vLLM.

All evaluation configurations and judge setups under the VLMEvalKit protocol are included to allow full benchmark reproducibility.

“The technology is ready to counter the industry’s obsession with data center centralization right now,” said Paolo Ardoino, CEO of Tether. “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. By putting these tools directly into the hands of developers, we are bypassing centralized gatekeepers and making powerful, private AI possible on the devices people already own.”

About QVAC and Tether AI Research

QVAC is Tether Data’s advanced AI research initiative dedicated to building open, decentralized, and adaptive intelligence systems. Its mission is Local AI and Infinite Intelligence. It is guided by an uncompromising vision of a world where AI lives and learns on any device, empowering individuals and communities rather than concentrating power in corporate data centers.

About Tether Data

Tether Data, S.A. de C.V. (“Tether Data”) is part of Tether’s broader vision to advance freedom, transparency, and innovation through technology. Its mission is to enable people and organizations to connect and share information directly, without unnecessary intermediaries. By creating secure, peer-to-peer systems, Tether Data gives users greater control over their data, communications, and digital interactions. Tether Data aims to redefine how information flows across networks by replacing centralized models with decentralized infrastructure designed for privacy, efficiency, and resilience. The company’s goal is to make global connectivity faster, safer, and more private, empowering individuals and institutions alike to exchange information freely and securely. References in this press release to Tether and Tether AI Research are references to Tether Data.