[Tether Open-Sources Edge-Side Vision-Language Model VisionPsy-Nano] BlockBeats News, July 29, according to the Tether blog, Tether Data's AI research project QVAC has open-sourced VisionPsy-Nano, a compact vision-language model designed for edge-side and edge deployment. With approximately 460 million parameters, it brings multimodal understanding capabilities—previously achievable only with cloud-based infrastructure—directly to mobile devices. Among all evaluated edge-side vision-language models with fewer than 500 million parameters, this model achieved the highest overall normalized score of 62.3, outperforming competing models such as Liquid AI and Hugging Face in 16 out of 17 benchmark tests. Tether has released two versions of the model, including the Flash version, which is optimized for low latency. The Flash version significantly improves inference speed while retaining approximately 99% of the full model's quality. Both versions are open-sourced under the Apache 2.0 license. Tether CEO Paolo Ardoino stated that this demonstrates the feasibility of a local-first, efficient AI approach, enabling powerful and private AI to run on people's existing devices. [Original Link]