IndiaAI Mission Launches Indigenous Foundation Models EKA-1 and Dhīr
New homegrown AI models prioritize offline functionality and support for multiple Indian languages.
1Launch of Indigenous Models
The IndiaAI Mission, under the Ministry of Electronics and IT, officially released three new foundation models. EKA-1 was developed by Soket Labs Technology and Research, while Dhīr 2B and Dhīr 5B were created by ZenteiQ AiTech Innovations. These releases represent a significant milestone in India's strategic push to build sovereign artificial intelligence capabilities.
2Government Infrastructure Support
The development of these models was facilitated by the IndiaAI Innovation Centre pillar. Participating firms received access to subsidized national AI compute infrastructure and financial support for research expenses. This public-private collaboration model demonstrates how state-backed resources can effectively accelerate the creation of homegrown technology and reduce reliance on foreign AI platforms.
3EKA-1 Technical Specifications
EKA-1 is a sparse hybrid model featuring 26.3 billion parameters trained on 7.5 trillion tokens. It is specifically optimized for mathematics, coding, reasoning, and agentic tasks. The model supports all 22 scheduled Indian languages alongside English, making it a versatile tool for the country's diverse linguistic and digital ecosystem requirements.
4Banking and Security Focus
Soket Labs designed EKA-1 to address the specific needs of India's banking, financial services, and cybersecurity sectors. By tailoring the model to these high-stakes industries, the developers aim to provide secure, domain-specific solutions. This focus ensures the technology remains relevant to the evolving digital landscape and complex regulatory requirements of Indian institutions.
5Dhīr Series Capabilities
ZenteiQ's Dhīr 2B and Dhīr 5B models are open-source tools supporting English and ten Indian languages. These models are engineered for document-grounded automation and knowledge retrieval. They are particularly effective for citizen and enterprise services, providing reliable assistance even in environments where internet connectivity is limited or strictly controlled for security.
6Offline and On-Device Utility
A core feature of these models is their ability to perform inferencing locally on devices like smartphones. By removing the dependency on constant cloud connectivity, the models enhance user privacy and data security. This on-device deployment capability allows for AI access in remote or sensitive environments where data transmission must be minimized.
7Strategic Technological Autonomy
IT Secretary S Krishnan emphasized that these indigenous models strengthen India's technological and strategic autonomy. By fostering local innovation, the government aims to improve productivity across defense, education, and industrial sectors. The initiative seeks to create a sustainable ecosystem where researchers and innovators can experiment with homegrown AI tools via platforms like AIKosh.
8Future Deployment Prospects
The lightweight architecture of these models allows them to run on various computing platforms, including GPUs and TPUs. As these technologies become more accessible through national platforms, their integration into government and educational workflows is expected to grow. This shift marks a transition toward localized, secure, and highly specialized AI infrastructure for India.