Breaking

Tether open-sources offline TranslatePsy models for 19 African and 9 European languages

September 2, 2026 at 10:51 EDT

Foundation Models
Open Source

Tether AI Research has released open-source translation models that run privately on phones, laptops, and other edge devices, targeting Sub-Saharan African languages first and adding a compact European counterpart. The September 2, 2026 launch covers QVAC TranslatePsy-AfriSLM (19 African languages), QVAC TranslatePsy-AfriNano (8 African languages), and QVAC TranslatePsy-EuroNano (9 European languages)—28 languages in total, matching the widely reported count of African and European coverage.

September 2, 2026 · Tether AI Research

Offline TranslatePsy for African and European languages

Open-source translation models run on phones and laptops. AfriSLM covers 19 African languages; EuroNano covers 9 European languages and 90 directions via an English pivot.

African languages (AfriSLM)

European languages (EuroNano)

0.8B

Smallest AfriSLM; beats larger MT systems on cited tests

96%

Low-quality training data filtered

Storage: EuroNano vs Firefox offline pack

Smallest EuroNano deployment is 36MB versus 633MB for an equivalent Firefox offline setup (~94% less).

36MB

EuroNano smallest

633MB

Firefox equivalent

Takeaway: One compact multilingual stack replaces a much larger bilingual-pack footprint.

What developers welcome

Small specialized models (0.8B–4B, quantized) that can translate locally without a cloud hop—useful where connectivity and privacy matter.

Open questions

African-language wins are company-reported; independent tests, local-context quality, and community feedback still need to catch up. Health use is education-only with safeguards.

On-device path — data stays on the handset; no third-party cloud required for inference.

1 · Download weights → 2 · Run locally → 3 · Translate offline

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.