India has taken another step towards developing its domestic artificial intelligence ecosystem with the release of three indigenous AI foundation models under the IndiaAI Mission. The Ministry of Electronics and Information Technology (MeitY) unveiled EKA-1, Dhīr 2B and Dhīr 5B, developed by Indian technology companies Soket Labs Technology and Research and ZenteiQ AiTech Innovations.
The models were unveiled in New Delhi on November 3, 2026, in the presence of MeitY Secretary S. Krishnan and senior government officials. The initiative aims to strengthen India’s technological independence while providing AI systems designed for Indian languages, specialised business requirements, secure environments and applications that can operate without continuous internet connectivity.
IndiaAI Mission Introduces Three Homegrown AI Models
The three models were developed under the Foundation Models pillar of the IndiaAI Innovation Centre. The government supported the development process through subsidised access to national AI computing infrastructure and assistance with additional research expenses.
The models address different requirements within India’s AI ecosystem. While EKA-1 focuses on advanced reasoning, mathematics, coding and agentic applications, ZenteiQ’s Dhīr models are designed for lightweight multilingual assistance and secure enterprise deployment.
Here is a comparison of the three models:
| Feature | EKA-1 | Dhīr 2B | Dhīr 5B |
|---|---|---|---|
| Developer | Soket Labs | ZenteiQ | ZenteiQ |
| Model size | 26.3 billion parameters | 2 billion parameters | 5 billion parameters |
| Training data | Approx. 7.5 trillion tokens | Not disclosed | Not disclosed |
| Language support | 22 Indian languages + English | 10 Indian languages + English | 10 Indian languages + English |
| Primary focus | Math, coding, reasoning and agents | Multilingual assistance | Multilingual assistance |
| Deployment | Local and enterprise environments | On-device and on-premise | On-device and on-premise |
| Availability | Indigenous foundation model | Open source | Open source |
EKA-1: Soket’s 26.3 Billion Parameter AI Model
Soket Labs Technology and Research developed EKA-1, an indigenous foundation model featuring 26.3 billion parameters and trained on approximately 7.5 trillion tokens.
The model supports all 22 scheduled Indian languages alongside English, making it particularly relevant to India’s multilingual technology ecosystem.
Soket describes EKA-1 as an open, sparse hybrid language model designed to handle several demanding AI applications, including:
- Mathematics and complex problem-solving.
- Code generation and programming assistance.
- Logical reasoning and analytical tasks.
- Agentic applications that can perform multi-step tasks.
One notable capability is its potential to perform certain inference operations directly on local devices. This means some AI functionalities can work without requiring a continuous mobile data or Wi-Fi connection.
The company is also positioning EKA-1 for specialised applications in India’s banking, financial services and cybersecurity industries.
According to Soket CEO Abhishek Upperwal, the model has been developed to address sector-specific requirements and India’s evolving digital ecosystem.
ZenteiQ Introduces Dhīr 2B and Dhīr 5B
ZenteiQ AiTech Innovations introduced two open-source AI models, Dhīr 2B and Dhīr 5B, designed for multilingual assistance and secure deployment.
Unlike larger general-purpose models that typically depend on cloud infrastructure, the Dhīr models focus on lightweight deployment across different computing environments.
Both models support English and 10 Indian languages.
Their key capabilities include:
- Document-grounded automation.
- Enterprise knowledge retrieval.
- Multilingual AI assistance.
- Citizen-facing digital services.
- Offline-capable AI applications.
The models are designed to operate in secure, on-premise environments, making them relevant to organisations that need greater control over their data.
AI Models Designed to Work Without Internet Connectivity
One of the most important features of Dhīr 2B and Dhīr 5B is their ability to support local deployment.
ZenteiQ says the models can operate across different computing platforms, including graphics processing units (GPUs) and tensor processing units (TPUs).
Their lightweight architecture also enables deployment on mobile devices, including smartphones, without depending on cloud connectivity.
This could be particularly useful for government departments, defence organisations, educational institutions and industrial facilities that operate in locations with limited internet access or strict data-security requirements.
ZenteiQ CEO Prof. Sashikumar said local deployment allows users to access AI capabilities while keeping their data on their own devices, improving privacy, security and accessibility.
Why India Is Investing in Indigenous AI Models
The development of domestic AI models is becoming an important part of India’s broader technology strategy.
India currently relies heavily on international AI companies and infrastructure for advanced artificial intelligence capabilities. Developing indigenous foundation models could help reduce this dependence and create technology better suited to local languages, industries and public services.
The IndiaAI Mission was approved by the Union Cabinet on March 7, 2024, with the objective of building a comprehensive domestic AI ecosystem.
The mission operates across seven major pillars:
- IndiaAI Compute Capacity: Expanding access to computing infrastructure required for AI development.
- IndiaAI Innovation Centre: Supporting the development of indigenous foundation models.
- IndiaAI Datasets Platform (AIKosh): Improving access to datasets for AI research and development.
- IndiaAI Application Development Initiative: Encouraging AI applications that address real-world challenges.
- IndiaAI FutureSkills: Developing AI skills and strengthening the talent ecosystem.
- IndiaAI Startup Financing: Supporting startups working on AI technologies.
- Safe and Trusted AI: Promoting responsible and secure AI development.
The government has also highlighted AIKosh as a platform through which researchers, students, innovators and businesses can access these models and experiment with their capabilities.
What These Models Could Mean for India’s AI Industry
The launch of EKA-1 and the Dhīr models could create opportunities across several industries.
In banking and financial services, specialised AI models could support document processing, customer assistance, compliance workflows and cybersecurity applications.
In government, multilingual AI assistants could help improve access to digital services and automate document-heavy administrative processes.
Educational institutions could also benefit from AI tools that understand Indian languages and provide assistance without requiring continuous internet connectivity.
For enterprises, on-premise deployment could be particularly important because sensitive business information does not always need to be transferred to external cloud services.
However, practical adoption will depend on model performance, reliability, hardware requirements, independent evaluations and the availability of developer tools.
The announcement has not provided comparative benchmark results establishing how these models perform against leading international AI systems.
The Bigger Picture
India’s release of EKA-1, Dhīr 2B and Dhīr 5B reflects a broader ambition to move beyond being primarily a consumer of international AI technology and become a developer of indigenous AI infrastructure.
The initiative also highlights an important difference in India’s AI strategy: developing models that address local requirements rather than competing only on model size or general-purpose capabilities. Support for Indian languages, offline operation and secure enterprise deployment could help address challenges that larger international platforms may not prioritise.
Government-backed computing infrastructure and research funding could also make it easier for domestic startups to develop foundation models without bearing the entire infrastructure cost independently. Nevertheless, long-term success will depend on whether these models achieve strong performance, attract developers and find meaningful commercial and public-sector applications.
Looking Ahead
The next stage for Soket and ZenteiQ will involve improving their models, expanding real-world deployments and demonstrating their capabilities through practical applications. Wider accessibility through platforms such as AIKosh could help researchers, developers and businesses experiment with the technology and identify new use cases.
For India’s broader AI ecosystem, these releases represent progress towards technological self-reliance. However, competing with established global AI developers will require continuous investment in research, computing infrastructure, high-quality datasets and talent. The real measure of success will be how effectively these indigenous models translate into widely adopted products and services.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
