QVAC Blog

29 July 2026 — Tether AI Research is releasing VisionPsy-Nano, a family of ~460M-parameter vision-language models built to run on the device in your pocket. It leads its weight class on 16 of 17 benchmarks, beats models up to 2.3x its size on ScienceQA, instruction following and hallucination robustness, and the Flash variant reaches the first token in 0.3s on an iPhone 15. Open weights, Apache 2.0.

16 July 2026 — For people who have lost their voice to ALS, the words are still there. The connection to the world is what breaks, not the mind. BrainWhisperer, from Tether Evo, decodes attempted speech from the brain into text, and QVAC runs it fully on-device: more than 90% of words correct on real recordings, in under 2 GB, nothing leaving the machine. An early proof of concept, not a product.

16 June 2026 — We built an app that predicts football matches. The twist: the AI runs entirely on your own machine, it simulates each match 10,000 times, and it reasons out loud before committing to a scoreline. No cloud model, no API key, no per-token bill. You can watch it think. This is a write-up of how it […]

2 June 2026 — TurboQuant is a KV-cache quantization algorithm published by Google Research at ICLR 2026 (Zandieh et al.). It compresses the running context memory of a transformer LLM by up to 5x with nearly no accuracy loss across long-context benchmarks. QVAC SDK 0.12.0 integrates TurboQuant inside qvac-fabric-llm.cpp with a Vulkan backend.