Cohere launches a family of open multilingual models
Source: TechCrunch
Overview
Enterprise AI company Cohere launched a new family of multilingual models, called Tiny Aya, on the sidelines of the India AI Summit. The models are open‑weight—meaning their underlying code is publicly available for anyone to use and modify—support over 70 languages, and can run on everyday devices like laptops without requiring an internet connection.
Supported Languages
The base model focuses on South Asian languages, including:
- Bengali
- Hindi
- Punjabi
- Urdu
- Gujarati
- Tamil
- Telugu
- Marathi
Model Variants
| Variant | Focus | Description |
|---|---|---|
| TinyAya‑Global | Broad multilingual coverage | Fine‑tuned to better follow user commands for apps that need wide language support. |
| TinyAya‑Earth | African languages | Regional variant optimized for African language families. |
| TinyAya‑Fire | South Asian languages | Tailored for the South Asian linguistic landscape. |
| TinyAya‑Water | Asia Pacific, West Asia, Europe | Designed for a diverse set of languages across these regions. |
The base model contains 3.35 billion parameters, a measure of its size and complexity.
Technical Details
- Training hardware: A single cluster of 64 NVIDIA H100 GPUs.
- Compute efficiency: Trained using relatively modest computing resources, making the models ideal for researchers and developers targeting native‑language audiences.
- On‑device usage: Built with software optimizations that require less computing power than most comparable models, enabling offline translation and other edge‑device applications.
“This approach allows each model to develop stronger linguistic grounding and cultural nuance, creating systems that feel more natural and reliable for the communities they are meant to serve. At the same time, all Tiny Aya models retain broad multilingual coverage, making them flexible starting points for further adaptation and research.” – Cohere statement
Availability
- Model repositories: Hugging Face, Cohere Platform, Kaggle, and Ollama.
- Datasets: Training and evaluation datasets are released on Hugging Face.
- Future releases: Cohere plans to publish a technical report detailing its training methodology.
Visuals
Image Credit: Cohere
Image Credit: Cohere
Context
Cohere’s CEO, Aidan Gomez, indicated last year that the company intends to go public “soon.” According to CNBC, Cohere ended 2025 with $240 million in annual recurring revenue and 50 % quarter‑over‑quarter growth.