Abu Dhabi-based AI group G42 on Tuesday announced the release of NANDA 87B, an open-source Hindi-English large language model with 87 billion parameters, an upgrade to its earlier NANDA model.
NANDA 87B is accessible as an open-weight resource on the MBZUAI Hugging Face page, allowing creators, developers, and businesses to utilise and expand its features.
The model has been developed by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in collaboration with Inception, a G42 company, and chipmaker Cerebras.
Built on Llama-3.1 70B, NANDA 87B has been trained on more than 65 billion Hindi tokens using a Hindi-centric tokeniser to improve efficiency in training and inference.
“India deserves world-class technology that speaks its language. NANDA 87B is a major step in that direction,” said Manu Jain, chief executive of G42 India, adding that the model is intended to support innovation across education, entertainment and enterprise use cases in India’s AI ecosystem.
G42 said the model is designed to handle formal Hindi, casual speech and Hinglish, and performs tasks such as translation, summarisation, instruction following and transliteration. The company added that safety and cultural alignment were part of the model’s design to ensure responsible outputs.
Richard Morton, executive director at MBZUAI’s Institute of Foundation Models, said the release marked progress in expanding access to advanced language technology. “NANDA marks an important milestone in bringing high-quality, open-access language technology to one of the world’s largest linguistic communities,” he said
The model was trained on Condor Galaxy, an AI supercomputing system built by G42 and Cerebras.