TranslatePsy-AfriSLM

Tether AI Research released TranslatePsy-AfriSLM on September 2, 2026, a family of open-source machine translation models designed to run entirely on device. The release covers 19 Sub-Saharan African languages including Hausa, Amharic, Yoruba, Swahili, Igbo, Zulu, Somali, Oromo and Xhosa.

The family ships in 800M, 2B and 4B parameter sizes. The smallest 800M model is compact enough to run offline on a phone or laptop, keeping user data on device rather than sending it to remote servers.

According to Tether's research paper, accepted at EMNLP 2026, the 800M model outperformed far larger systems including Qwen3.5-122B-A10B, TranslateGemma-27B and NLLB-3.3B across FLORES-200, BOUQuET and SMOL benchmarks. The team reports it removed 96% of low-quality training material, building on about 215 million bidirectional training examples.