By Hamza Faisal
India has unveiled three homegrown artificial intelligence foundation models, including two designed to run entirely offline, in its latest push to build AI technology that speaks the country's languages and keeps sensitive data inside its borders. The models were released in New Delhi on Tuesday by the IndiaAI Mission, the government's flagship AI programme, in the presence of S. Krishnan, Secretary of the Ministry of Electronics and Information Technology.
The launch pairs two very different approaches to domestic AI. Soket Labs' EKA-1 is a large model aimed at complex enterprise work, while ZenteiQ's Dhīr 2B and Dhīr 5B are small, open-source models built to run on ordinary devices without a constant internet connection. All three were developed under the IndiaAI Mission's Innovation Centre, which supports Indian teams building foundation models with subsidised national computing infrastructure.
EKA-1: a large model for banking, cybersecurity and reasoning
EKA-1 is the heavyweight of the trio. Soket Labs describes it as an open, sparse hybrid language model with 26.3 billion parameters, trained on approximately 7.5 trillion tokens. It covers all 22 languages listed in the Eighth Schedule of India's Constitution alongside English, and it is pitched at banking, financial services and cybersecurity, as well as mathematics, code and reasoning tasks.
Notably, EKA-1 can also perform some inference directly on a local device, such as a smartphone, without mobile data or Wi-Fi — a feature that could matter for services in areas where connectivity is patchy, or where sending data to a remote cloud is undesirable. The launch announcement does not, however, establish that EKA-1 has already been adopted by banks or deployed in operational cyber-defence systems.
Dhīr 2B and 5B: small models that work without the cloud
ZenteiQ's Dhīr models take the opposite approach. Dhīr 2B and Dhīr 5B are open-source, lightweight multilingual assistants supporting English and 10 Indian languages, designed for document-grounded automation, knowledge retrieval and citizen or enterprise services across defence, government, industrial and educational sectors.
ZenteiQ, a startup born at the Indian Institute of Science in Bengaluru and led by founder Prof. Sashikumaar Ganesan, says the models are compact enough to run fully offline on laptops and edge devices, and can be deployed on an organisation's own infrastructure as secure on-premise assistants. Dhīr 2B is a dense model with 2.03 billion parameters, while Dhīr 5B uses a mixture-of-experts design with 5.09 billion total parameters, of which 1.79 billion are active at a time. Both were pretrained from scratch rather than adapted from existing model weights.
What the launch means — and what is still unproven
The models add to earlier foundation models released under the IndiaAI Mission, including offerings from Sarvam AI, BharatGen and Gnani AI, as New Delhi tries to reduce dependence on AI technology built and controlled overseas. The mission itself was approved by India's Union Cabinet in March 2024 as part of a broader effort to expand computing infrastructure, AI research and startup development.
Readers should keep the launch in proportion. The government has not supplied comprehensive independent benchmark results showing how the three models compare with established international alternatives, and detailed deployment figures are not yet available. Parameter counts and training-token totals are useful context, not proof of quality — and offline capability says nothing about accuracy in a particular language. What is clear is the direction of travel: models built in India, for Indian languages, for settings where sending data to a distant cloud is not always desirable or possible.



