Atrás

Cohere Unveils Open-Weight Tiny Aya Multilingual Model Family

Launch Overview

Enterprise artificial‑intelligence company Cohere announced the release of a new family of multilingual models called Tiny Aya. The announcement took place alongside the ongoing India AI Summit and was presented by Cohere Labs, the company’s research division. The models are open‑weight, meaning the underlying code is publicly available for anyone to use, modify, or redistribute.

Model Capabilities

The Tiny Aya family supports more than 70 languages and is engineered to run on everyday devices such as laptops, eliminating the need for a constant internet connection. The base model contains 3.35 billion parameters, a size that balances linguistic capability with modest computational demands. The suite includes a globally‑focused variant, TinyAya‑Global, fine‑tuned to follow user commands across a broad language set, as well as three regional versions: TinyAya‑Earth for African languages, TinyAya‑Fire for South Asian languages—including Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, and Marathi—and TinyAya‑Water for languages spoken in Asia‑Pacific, West Asia, and Europe.

Technical Details

All Tiny Aya models were trained on a single cluster of 64 H100 GPUs, a high‑performance chip from Nvidia, using relatively modest computing resources. Cohere designed the underlying software to suit on‑device usage, requiring less computing power than most comparable models. This architecture enables offline translation and other language‑intensive tasks without reliance on cloud services, a benefit highlighted for linguistically diverse markets like India.

Availability and Ecosystem

The models are publicly available on the HuggingFace platform and the Cohere Platform. Developers can also download them from Kaggle and Ollama for local deployment. Cohere is releasing the training and evaluation datasets on HuggingFace and plans to issue a technical report that details the training methodology. These resources aim to help researchers and developers build applications that serve native‑language audiences.

Business Context

During the announcement, Cohere’s CEO Aidan Gomez reiterated the company’s intention to go public “soon.” Financial reporting cited by CNBC indicated that Cohere ended 2025 with $240 million in annual recurring revenue and experienced 50 % growth quarter‑over‑quarter throughout the year. The launch of Tiny Aya aligns with Cohere’s broader strategy to provide accessible, high‑quality language models for developers and enterprises worldwide.

Usado: News Factory APP - descubrimiento de noticias y automatización - ChatGPT para Empresas

Source: TechCrunch

También disponible en: