Offline, on-device translation for underserved communities Stablecoin issuer @Tether has released a family of open-source neural machine translation models designed to run entirely offline on
Stablecoin issuer @Tether has released a family of open-source neural machine translation models designed to run entirely offline on phones and laptops, covering 19 African languages spoken by roughly half the continent's population. A companion release, TranslatePsy-EuroNano, extends the same technology to nine European languages.
The model family is called TranslatePsy-AfriSLM. It covers 19 Sub-Saharan African languages and was trained on a dataset of roughly 215 million bidirectional examples paired with English. Crucially, the models are compact, ranging from 21 to 35 MB per language pair, small enough to run directly on a phone without connecting to a server. That design choice reflects a deliberate push for privacy and accessibility in regions where cloud infrastructure is limited.
Alongside the African language models, Tether released the QVAC SDK, a software development kit that bundles various AI models into a unified deployment framework, with European language support coming through this SDK rather than through separate dedicated datasets.
Tether says the translation models are intended to pair with its existing medical AI model, enabling health information to be delivered in local languages. Education content and farming guidance are also listed as target applications. All models are free to download.
Tether's stablecoin sees significant adoption across Africa, Southeast Asia, and Latin America, regions where local currencies are volatile, banking infrastructure is thin, and where many of the world's low-resource languages are spoken. The AI push appears to extend that footprint beyond finance.
The research underpinning the release has been accepted at the EMNLP 2026 conference. The accompanying paper is available on arXiv (arXiv:2608.18655), and the code and models are documented on GitHub under Tether's AI research organisation.
The broader challenge is significant. Africa is home to over 2,000 native languages, and although many have millions of speakers, most remain low-resource languages with limited AI support. Tether's release is a targeted attempt to close part of that gap.
Sources:Crypto Briefing: Tether releases open-source AI translation models for African and European languagesTether official website: TranslatePsy-AfriSLM announcementarXiv: AfriNLLB, research on efficient translation models for African languages
