Tiny Aya in the Wild: What People Are Building With Open Multilingual AI
Cohere, Tuesday, July 14th, 2026
Cohere Labs showcases 13 projects built on its open-weight Tiny Aya multilingual model across education, safety, and on-device AI.
Cohere Labs recaps the Expedition Tiny Aya research program, in which global teams built applications on the open-weight multilingual model.
Thirteen projects are grouped into four themes: education and learning, multilingual safety, accessibility and deployment, and language understanding. Safety researchers found that model behavior shifts dramatically under code-mixed prompts, exposing gaps in evaluation frameworks that test languages in isolation.
Other teams developed compression techniques, synthetic data pipelines, and offline on-device assistants, showing multilingual AI can run efficiently on phones.
The core Tiny Aya research was accepted to COLM 2026, and the initiative demonstrates how open models plus mentorship drive innovation for underrepresented languages.