TechFlow reports that on September 2, according to Tether's official website, Tether AI Research announced the release of the open-source QVAC TranslatePsy series of AI translation models, supporting 19 African languages (TranslatePsy-AfriSLM), 8 African languages (TranslatePsy-AfriNano), and 9 European languages (TranslatePsy-EuroNano). The models can run locally and offline on end-user devices such as smartphones and laptops without requiring an internet connection, and data is not uploaded to third-party cloud servers.
Among these, the smallest version of the AfriSLM model, comprising only 800 million parameters, outperforms larger models such as Qwen3.5-122B-A10B and TranslateGemma-27B across three translation benchmarks: FLORES-200, BOUQuET, and SMOL. The smallest version of EuroNano requires only 36MB of storage space, representing an approximate 94% reduction compared to Firefox's offline translation configuration. The model series is now publicly available for download on Hugging Face, and the associated research has been accepted at EMNLP 2026.





