Research · Models

Models that learn for the world people live in

We train and release models that advance useful, accessible intelligence, starting with the languages we work in. Browse current open weights below.

How we approach models

Our research goals push toward human-level intelligence that is efficient, adaptive, and able to reason, not language alone. Every model we ship should help people collaborate with AI in their own languages and constraints.

Efficient by intent

Architectures and training recipes that work with limited data and compute, so intelligence can travel.

Languages people speak

Speech, translation, and generation for Swahili, Sukuma, Runyambo, and more languages we work in.

Open for builders

Open weights and clear notes so others can try, fine-tune, and run models themselves.

Toward fuller intelligence

Language is necessary but not sufficient. We keep building capabilities that support real collaboration and impact.

Current models

Speech to text (STT), text to speech (TTS), machine translation (MT), and more.

12
Models
3
Languages
12
Open weights

Showing 12 models