Abu Dhabi’s G42 has made a major upgrade to Llama-3-Nanda (NANDA), its open-source Hindi-English large language model (LLM), which now features 87 billion parameters and sets a new benchmark in language AI tailored for Hindi speakers.
It makes NANDA the largest and one of the most capable Hindi-centric models available in open weights.
Built upon Llama-3.1 70B, NANDA 87B has been trained on a curated Hindi-English dataset with over 65 billion Hindi tokens. A custom Hindi-centric tokeniser boosts efficiency, reducing both training and inference time.
NANDA is developed by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in collaboration with Inception, a G42 company, and Cerebras, makers of the world’s fastest AI inference.
Engineered for real-world use, NANDA is fluent in formal Hindi (Devanagari script), casual speech, and Hinglish. It delivers strong performance across translation, summarisation, instruction-following, and transliteration tasks. Safety and cultural alignment are core to its design, enabling NANDA to generate context-aware, responsible responses.
NANDA 87B sets new benchmark
With over 600 million Hindi speakers and one of the world’s fastest-growing digital economies, India represents a crucial market for regional AI innovation. With over 80 per cent of new internet users preferring local languages, models like NANDA can play a pivotal role in bridging digital divides.
Manu Jain, CEO of G42 India, said: “India deserves world-class technology that speaks its language. NANDA 87B is a major step in that direction, after our first NANDA model was announced last year.
“As we continue to scale our operations across the country, this model opens doors for more inclusive innovation in education, entertainment, enterprise, and beyond. This upgrade reflects G42’s deep commitment to building AI solutions that serve India’s vibrant AI ecosystem.”
Richard Morton, Executive Director, Institute of Foundation Models, MBZUAI, added: “At MBZUAI, our mission is to advance AI in ways that deliver broad, positive impact for society. NANDA marks an important milestone in bringing high-quality, open-access language technology to one of the world’s largest and most dynamic linguistic communities.
“Through our collaboration with G42 and Cerebras, we are underscoring the value of culturally aligned and inclusive AI research — work that supports underserved languages and expands access to advanced AI capabilities for hundreds of millions of Hindi speakers worldwide.”
The model was trained on Condor Galaxy, one of the world’s most powerful AI supercomputers for training and inferencing, built by G42 and Cerebras. A new Hindi LLM is now available as an open-weight model on MBZUAI Hugging Face page, enabling creators, developers, and enterprises to explore and build upon its capabilities.
Follow us on
