NEW DELHI: Three indigenous artificial intelligence (AI) models launched under the Centre’s IndiaAI Mission are being developed for applications ranging from offline access on mobile phones to secure use in critical sectors such as defence, cybersecurity and banking.
The models — EKA-1, Dhīr 2B and Dhīr 5B — were launched by the Union Ministry of Electronics and Information Technology (MeitY) on Tuesday under the Foundation Model initiative of the IndiaAI Mission. Developed by Bengaluru-based Socket AI and ZenteiQ, the models aim to reduce India’s dependence on foreign AI technologies and cater to the country’s specific needs.
ZenteiQ’s Dhīr 2B and Dhīr 5B are compact AI models that can operate directly on mobile phones without requiring an internet connection. This is expected to make AI services accessible to users in areas with limited internet connectivity.
ZenteiQ founder and chief scientist Professor Sashikumaar Ganesan said users would be able to access the models directly on their mobile devices after downloading them. “Once downloaded, all the configuration automatically takes place, then the model will be sitting in the user’s mobile itself,” he said.
#WATCH | Delhi: On public release of Indigenous AI foundation models Dhīr 2b and Dhīr 5b, CEO of Zenteiq Professor Sashikumaar Ganesan says, “We have launched two small models. We call them compact models…what we have launched is a two-billion and then five-billion parameter model…we can run these models on our mobile device itself… users can get AI solutions in their hands without even going to the cloud, number one…the mobile app that we are coming up with, that is also going to be in the Play Store as well as the iOS so that the user can download…it will be free…another model that we are going to launch is the 9-billion parameter model…”
— ANI (@ANI) November 3, 2026
Ganesan said the company was developing a free mobile application for both Android and iOS users. The model weights are already available through AIKosh, while the company is planning to launch a 9-billion-parameter model in December for researchers and engineers working on specialised problems.
Meanwhile, Socket AI’s EKA-1 is a 26-billion-parameter model trained on 7.5 trillion tokens. It is designed to perform tasks related to mathematics, coding and reasoning and can be deployed on private servers, private cloud systems and environments without internet connectivity.
Socket AI CEO Abhishek Upperwal said the model was developed to reduce dependence on foreign AI technologies, particularly in sectors where data security and control are important. “We want to remove the reliance of foreign models and we want to make an indigenous model which can work in all our critical sectors,” he said, referring to areas such as defence, cybersecurity, banking and finance.
#WATCH | Delhi: CEO and founder of Soket AI Abhishek Upperwal says, “…sectors such as defence, cybersecurity, and banking and finance rely heavily on mathematics and coding. Our goal is to reduce India’s dependence on foreign models by developing indigenous alternatives that can function across all our critical sectors. This ensures that sensitive data—such as that from the banking and finance sectors—does not leave the country, preventing potential data leaks. The model developed by our team can be deployed on-premises or within data centres located in India. It can also be fine-tuned to suit your specific requirements. You have complete flexibility regarding how you use the model and where your data is stored; there is no reliance on foreign models or technologies…”
— ANI (@ANI) November 3, 2026
Socket AI Chief Science Officer and co-founder Mayank Singh said the company had developed the model’s data preparation, training and evaluation processes entirely in India. The company plans to make the model weights openly available within a month, with initial access being provided through a waitlist.
#WATCH | Delhi: Chief Science Officer and Co-Founder, Soket AI, Mayank singh, says, “In my view, this is a major achievement because, as is generally known, many models in India have not been trained from the very beginning. Typically, open models are made available, which are then fine-tuned. In contrast, this model was trained entirely from scratch…”
— ANI (@ANI) November 3, 2026
At the launch, MeitY Secretary S. Krishnan said the indigenous models could strengthen India’s strategic autonomy in artificial intelligence. He particularly highlighted their potential for enterprise applications in secure, air-gapped environments, where systems operate without a connection to external networks.
The launch is part of India’s broader efforts to build domestic AI capabilities and develop technologies that can be deployed across different sectors while ensuring greater control over data and digital infrastructure. (ANI)
