Platform
Models
Featured
Deploy
By Industry
Featured
Explorations
Initiatives
Resources
MEET AYA
Aya, a global open-science initiative from Cohere Labs, unites researchers to push the frontier of multilingual AI, bridging gaps between people and cultures worldwide.
NEW RELEASE
A compact multilingual AI model family that runs locally on any device. Our 3.35B-parameter base model supports 70+ languages with specialized variants for different regions, delivering strong performance without cloud dependency. Choose from globally balanced or region-specialized models designed for real-world scenarios.
Aya Vision is a research model advancing in multilingual multimodal AI through innovative synthetic data generation, cross-modal model merging, and a comprehensive benchmark suite. It achieves state-of-the-art performance across 23 languages, surpassing larger models while efficiently addressing data scarcity and catastrophic forgetting by reducing computational overhead up to 40% via optimized training techniques.
Aya Expanse redefines multilingual AI as a research model mastering 101 languages through innovative instruction tuning and cross-lingual transfer techniques. By combining a curated open-source dataset with compute-efficient pretraining, it achieves unparalleled performance across low- and high-resource languages while reducing infrastructure costs by up to 30%—setting a new benchmark for scalable, inclusive language modeling.
Aya 101 is a groundbreaking research model delivering instruction-tuned proficiency across 101 languages, developed through a global collaborative effort involving 3,000+ researchers. As the cornerstone of Cohere Labs’ multilingual initiative, it establishes a new standard for inclusive AI by achieving state-of-the-art performance on diverse benchmarks while maintaining efficiency across low- and high-resource languages—all released as an open-source framework to accelerate equitable language technology worldwide.
Under the hood
Open collaboration and transparency
Aya’s models and datasets are openly licensed, fostering global collaboration and enabling researchers to audit, extend, and innovate responsibly. This transparency accelerates collective progress in multilingual AI while ensuring accessibility for all.
Resource-efficient innovation
Aya leverages modular training and lean model designs to minimize computational costs without sacrificing performance. This efficiency democratizes advanced AI development, making cutting-edge research feasible for diverse communities.
High quality datasets
Aya prioritizes diverse data and adaptive architectures to deliver equitable performance across high- and low-resource languages. Its instruction-tuned pipelines ensure seamless cross-lingual task generalization, breaking barriers in global AI utility.
Stanford University HAI
For “Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning”
ACL 2024
For “Aya Model: An Instruction Finetuned Open-Access Multilingual Language Model”
Run multilingual AI locally with 3.35B-parameter models covering 70+ languages and specialized regional performance.
Aya began as the largest open science collaboration in ML, bringing together a community of researchers from around the world to create a powerful foundation for future innovation. This initial effort laid the groundwork for subsequent research initiatives and the development of additional models, pushing the boundaries of what is possible and expanding the world that AI sees.
languages contributed
independent researchers
countries
models
Feb 13, 2024
