Tether opens a 460M-parameter edge vision model

Tether’s QVAC released VisionPsy Nano, a 460-million-parameter vision-language model that led 16 of 17 vendor-run benchmarks against tested sub-0.5B rivals. Its Flash variant retains about 99% of the base model’s reported quality while producing a first token 19–23 times faster on three Android phones and up to 36 times faster on an iPhone 15; both variants ship as Apache 2.0 weights with Transformers, GGUF, and vLLM paths. For builders tracking the control promised by large open-weight systems, this is the opposite bet: private OCR and visual assistance may fit on the device before they need frontier-scale reasoning.