India's Indigenous AI Models: EKA-1 and Dhīr Explained
Artificial intelligence is increasingly becoming a crucial element in economic growth, education, healthcare, financial services, and public administration. Nonetheless, the global AI sector is largely controlled by a limited number of international tech firms that create models and infrastructure. For India, developing its own artificial intelligence capabilities is not only about technological advancement; it also concerns accessibility, data protection, and enduring digital autonomy.
On 3 November 2026, the Ministry of Electronics and Information Technology (MeitY), as part of the IndiaAI Mission, declared the launch of three homegrown foundation models: EKA-1 from Soket Labs and Dhīr 2B and Dhīr 5B from ZenteiQ AiTech Innovations. These models embody various strategies for India's AI goals, encompassing multilingual reasoning and business applications, to lightweight assistants capable of functioning in offline settings.
What Are India's Indigenous AI Models?
Indigenous AI models are AI systems created by Indian firms, research groups, or institutions to meet local technological needs. Enhancing their development can boost assistance for Indian languages, local datasets, regional applications, and particular industry needs. In contrast to applications that merely incorporate a pre-existing global AI model, foundation models offer the essential functionalities that developers can customize to build specialized applications.
The recent launch of EKA-1 and Dhīr shows that India is investigating various AI systems instead of depending on one model for all needs. EKA-1 emphasizes advanced reasoning, mathematics, and coding, whereas the Dhīr models concentrate on smaller-scale implementation, multilingual support, and local data handling.
EKA-1: A Foundation Model Designed for Complex Tasks
Created by Soket Labs Technology and Research Private Limited, EKA-1 is a foundational model with 26.3 billion parameters that has been trained on roughly 7.5 trillion tokens.
It is created as a sparse, open hybrid language model and accommodates all 22 scheduled languages of India in addition to English. Its envisioned functions encompass mathematics, coding, logical reasoning, and autonomous applications, where an AI system is capable of performing multi-step tasks.
A key feature of EKA-1 is its applicability to specialized business applications. Soket's leadership indicates that the model has been crafted with consideration for the banking, financial services, and cybersecurity needs of India.
These abilities could aid in applications related to financial data analysis, technical support, security processes, and organizational efficiency. Its multilingual features could assist developers in building tools that function effectively in India's varied linguistic landscape.
Nonetheless, these represent intended uses rather than evidence that every capability has already been widely implemented commercially. Objective performance assessments and practical testing will be crucial in assessing how well EKA-1 operates compared to recognized global standards.
Dhīr 2B and Dhīr 5B: Smaller Models for Local and Offline AI
EKA-1 is tailored for rigorous applications, whereas ZenteiQ's Dhīr 2B and Dhīr 5B models emphasize accessibility and ease of deployment.
The two open-source models provide support for English and 10 languages from India. They are intended for automation based on documents, knowledge acquisition, and multilingual support in areas like government, defense, education, and industry.
A key characteristic is their capability to enable on-premises and offline deployment. This implies that organizations can operate the models in their own managed computing settings instead of relying solely on external cloud services.ZenteiQ has stated that the lightweight architecture enables deployment on compatible mobile devices. This local processing could be beneficial in regions with poor internet access or situations where confidential data needs to stay within an organization's infrastructure.
An educational organization might utilize a locally installed assistant to aid students in finding learning resources in Indian languages. A government agency might utilize a document-driven assistant to access data from internal records without transferring those documents to an outside cloud service.
These abilities might be especially pertinent for organizations managing sensitive data. Nonetheless, local deployment does not inherently ensure security; suitable access controls, encryption, and device safeguarding are still required.
Why Indian-Language Support Matters
India's varied languages pose a significant obstacle for the implementation of AI. Numerous individuals express themselves more easily in their native languages compared to English, yet technical resources and digital services vary in accessibility among different languages.
AI models that facilitate Indian languages have the potential to enhance access to education, digital governance, healthcare data, and financial services. They can assist developers in building applications for users who lack access to English-centered technology.EKA-1's assistance for all 22 scheduled Indian languages alongside English signifies a move towards wider linguistic inclusion. Dhīr's backing for 10 Indian languages provides an alternative method using smaller models that can be implemented nearer to users.
However, language support should not be assessed solely based on the count of languages provided. The precision of translation, cultural relevance, regional lexicon, dialect identification, and effectiveness in daily interactions will ascertain if these models offer worthwhile advantages.
The Role of the IndiaAI Mission
The creation of these models is linked to the Government of India's IndiaAI Mission, sanctioned by the Union Cabinet on 7 March 2024.
With a sanctioned budget of ₹10,371.92 crore over five years, the initiative aims to build a broader national AI ecosystem through seven foundational elements: computing power, innovation hubs, datasets, application creation, skill development, startup funding, and secure, reliable AI.
Soket and ZenteiQ were chosen as part of the foundation-model initiative by the IndiaAI Innovation Centre. The government granted access to subsidized national AI computing resources and assistance for extra research costs.
Platforms like AIKosh aim to assist researchers, students, startups, and developers in finding and utilizing datasets, models, and various AI resources. This assistance is crucial since training foundational models demands significant computing power, technical skills, high-quality datasets, and ongoing financial investment. Public backing can lower entry barriers for Indian startups that might otherwise find it difficult to compete with firms having significantly greater financial and computational assets.
Why Indigenous AI Matters for India's Technological Independence
Creating indigenous AI models can enhance India's influence over crucial aspects of its digital landscape. It may generate chances for national research, stimulate regional innovation, and lessen reliance on a small group of external AI suppliers.
It might also assist organizations in creating tailored systems for Indian legislation, languages, industries, and public sector needs. Models that facilitate local or offline deployment may offer greater adaptability for organizations managing sensitive data.
Yet, technological self-sufficiency does not imply that all parts need to be produced in the home country. Computing hardware, semiconductor supply chains, cloud infrastructure, and specialized development tools continue to be vital components of the larger AI ecosystem. Developing sustainable national capabilities will necessitate ongoing investment in these sectors.
India needs to tackle issues related to model reliability, hallucinations, data quality, bias, cybersecurity, computing expenses, and the availability of qualified professionals. Independent benchmarks, clear documentation, and responsible implementation will be crucial for fostering trust in indigenous models.
Conclusion
The launch of EKA-1 along with Dhīr 2B and 5B signifies a significant milestone in India's pursuit of creating indigenous artificial intelligence. Their emphasis on multilingual features, specialized applications, and adaptable deployment showcases the nation's varied technological needs. The true indicator of success will be their efficacy in real-world applications, availability for developers, and capacity to consistently tackle local issues. India's AI goals should extend beyond just developing models; they need to foster a sustainable ecosystem where Indian creativity, skills, infrastructure, and languages influence the future of artificial intelligence.
Sources
Press Information Bureau, Ministry of Electronics and Information Technology. IndiaAI Mission Releases Indigenous Foundation Models EKA-1 by Soket and Dhīr 2B & Dhīr 5B by ZenteiQ Under the IndiaAI Innovation Centre Pillar. 3 November 2026. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2332315
Press Information Bureau, Government of India. Government Expands Sovereign AI Infrastructure Through IndiaAI Mission and Semiconductor Initiatives. 6 August 2026. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2295477
Press Information Bureau, Government of India. IndiaAI Mission: Sovereign AI Computing Capacity and Indigenous Foundation Models. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2298788
Ministry of Electronics and Information Technology. IndiaAI Mission. https://indiaai.gov.in/
TVR Perspective
India's development of indigenous AI models represents an important step towards building technological capabilities suited to its own languages, industries and public needs. However, the long-term value of these models will depend on more than their launch. Reliability, accessibility, responsible deployment, data protection and real-world performance will determine whether homegrown AI can become a meaningful part of India's digital future.
MAYURI NILAWAR
TVR Contributor
Category: Technology & AI
Writer ID: TVR-W002-TECH
TVR contributor covering technology, artificial intelligence and digital developments.
Reviewed and edited by the TVR Editorial Team.

Comments