Enterprise AI company Cohere has launched a new family of multilingual models on the occasion of the ongoing India AI Summit. These models, called Tiny Aya, are open weight. This means the underlying code is open for anyone to use and modify, supports over 70 languages, and can run on everyday devices like laptops without the need for an internet connection.
The model, launched by the company’s research arm Cohere Labs, supports South Asian languages such as Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu and Marathi.
The base model contains 3.35 billion parameters, which is a measure of its size and complexity. Cohere also released TinyAya-Global, a tweaked version that better follows user commands for apps that require extensive language support. Regional variations complete this family. TinyAya-Earth for African languages. TinyAya-Fire for South Asian languages. TinyAya-Water for Asia Pacific, West Asia and Europe.
“This approach allows each model to develop stronger linguistic foundations and cultural nuances, creating a system that is more natural and reliable for the communities it serves. At the same time, all Tinyaya models have extensive multilingual coverage, providing a flexible starting point for further adaptation and research,” the company said in a statement.
Kohia noted that these models were trained on a single cluster of 64 H100 GPUs (a type of Nvidia’s high-performance chips) using relatively modest computing sources, making them ideal for researchers and developers building apps for native-speaking audiences. These models can run directly on the device, so developers can use them to power offline translation. The company noted that it builds the underlying software for use on the device and requires less computing power than most comparable models.
In a linguistically diverse country like India, this kind of offline-enabled functionality enables a variety of applications and use cases without the need for constant access to the internet.
The model is available on the HuggingFace and Cohere platforms, popular platforms for sharing and testing AI models. Developers can download and deploy locally on HuggingFace, Kaggle, and Ollama. The company is also releasing training and evaluation datasets on HuggingFace and plans to release a technical report detailing its training methodology.
tech crunch event
boston, massachusetts
|
June 23, 2026
The company’s CEO, Aidan Gomez, said last year that the company plans to go public “soon.” According to CNBC, the company ended 2025 on a strong note, posting $240 million in annual recurring revenue and 50% quarter-over-quarter growth throughout the year.
