Abu Dhabi–based G42 has launched NANDA 87B, an open-weight Hindi–English large language model.
Abu Dhabi–based AI group G42 has released NANDA 87B, an open source Hindi–English large language model with 87 billion parameters, strengthening the availability of high-scale open AI for one of the world’s largest language communities.
NANDA 87B is available as an open-weight model on the MBZUAI Hugging Face page, allowing developers, researchers, startups, and enterprises to use, modify, and extend the model without vendor lock-in. The release marks a major upgrade over G42’s earlier NANDA model, significantly increasing both scale and capability.
The model is designed specifically for Hindi, English, and Hinglish, and has been trained on more than 65 billion Hindi tokens using a Hindi-centric tokeniser to improve training efficiency and inference performance. Built on Llama-3.1 70B, NANDA 87B extends an established open-model foundation rather than relying on closed architectures.
Development was led by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in collaboration with Inception, a G42 company, and chipmaker Cerebras. Training was carried out on Condor Galaxy, the AI supercomputing system built by G42 and Cerebras.
“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, noting that the model is intended to support education, entertainment, and enterprise use cases across India’s AI ecosystem.
Richard Morton, Executive Director at MBZUAI’s Institute of Foundation Models, said, “NANDA marks an important milestone in bringing high-quality, open-access language technology to one of the world’s largest linguistic communities.”
NANDA 87B supports translation, summarisation, instruction following, and transliteration, while embedding safety and cultural alignment to enable responsible open-source AI deployment.
