Cohere Aya
By Cohere For AI
Cohere Aya refers to the Aya family of open-weight multilingual language models released by Cohere For AI, Cohere's research lab, spanning the original Aya 101 model and the newer Aya Expanse and Aya Vision generations.
Definition
Cohere Aya refers to the Aya family of open-weight multilingual language models released by Cohere For AI, Cohere's research lab, spanning the original Aya 101 model and the newer Aya Expanse and Aya Vision generations.
Overview
Cohere Aya is the branding used to describe the Aya open-model initiative from Cohere For AI, distinguishing it from Cohere's commercial Command model line. While Command models are Cohere's proprietary, API-served flagship products optimized for enterprise retrieval-augmented generation and tool use, Aya is an openly released research and community initiative centered on multilingual coverage. The two projects share an underlying organization and some research investment but serve different purposes: Aya is meant to advance open multilingual NLP research and provide accessible high-quality open-weight models, particularly for languages underserved by mainstream commercial LLMs. The Aya lineage began with the Aya Dataset and Aya Collection — large-scale, crowdsourced multilingual instruction-tuning datasets spanning 101 languages, produced through a global open-science collaboration — and the original Aya 101 model, a 13B parameter model fine-tuned from mT5. Cohere For AI followed with Aya Expanse (8B and 32B open-weight models built on more modern architectures with improved quality via techniques like data arbitrage and model merging) and Aya Vision, which added multimodal image-understanding capability across many supported languages. Because Aya models and datasets are released openly (typically via Hugging Face), they have become a common reference point and building block for academic and community researchers working on multilingual NLP, low-resource language modeling, and cross-lingual evaluation, distinguishing Cohere's public research contributions from its commercial product surface built around the Command model family.
Key Features
- Open-weight multilingual model family distinct from Cohere's commercial Command line
- Built on openly released Aya Dataset and Aya Collection spanning 101 languages
- Aya Expanse (8B/32B) generation improved quality via data arbitrage and model merging
- Aya Vision adds multimodal image understanding across many languages
- Produced through Cohere For AI's global open-science research collaboration
- Freely available via Hugging Face for research and community use