Cohere launches a family of multilingual open templates

Enterprise AI company Cohere has launched a new family of multilingual models on the sidelines of the ongoing India AI Summit. The models, called Tiny Aya, are open-weight — meaning their underlying code is publicly available for anyone to use and modify — supports more than 70 languages, and can run on everyday devices like laptops without requiring an internet connection.

The model, launched by Cohere Labs, the company’s research arm, supports South Asian languages ​​such as Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu and Marathi.

The basic model contains 3.35 billion parameters, which is a measure of its size and complexity. Cohere also launched TinyAya-Global, a version tuned to better follow user commands, for applications that require broad language support. Regional variants surround the family: TinyAya-Earth for African languages; TinyAya-Fire for South Asian languages; and TinyAya-Water for Asia Pacific, West Asia and Europe.

“This approach allows each model to develop stronger linguistic foundations and cultural nuances, creating systems that feel more natural and reliable for the communities they are meant to serve. At the same time, all Tiny Aya models maintain broad multilingual coverage, making them flexible starting points for further adaptation and research,” the company said in a statement.

These models, trained on a single cluster of 64 H100 GPUs (a type of high-power chip from Nvidia) using relatively modest compute resources, are ideal for researchers and developers building applications for audiences that speak native languages, Cowher noted. The models are able to run directly on devices, so developers can use them to run translations offline. The company indicated that it designed its basic software to suit use on the device, which requires less computing power than most similar models.

In linguistically diverse countries like India, this kind of convenient offline capabilities can open up a variety of applications and use cases without the need for constant Internet access.

The models are available on HuggingFace, the popular platform for sharing and testing AI models, and Cohere. Developers can download it to HuggingFace, Kaggle, and Ollama for on-premises deployment. The company is also releasing training and evaluation datasets on HuggingFace and plans to release a technical report detailing its training methodology.

The startup’s CEO, Aidan Gomez, said last year that the company plans to go public “soon.” According to CNBC, the company ended 2025 on a high note, recording $240 million in annual recurring revenue, with 50% growth quarter-over-quarter over the year.