India Launches 3 Homegrown AI Models: EKA-1 and Dhīr

India Launches 3 Homegrown AI Models: EKA-1, Dhīr 2B and Dhīr 5B Explained

India's artificial intelligence industry has reached another important milestone. The Government of India has released three indigenous AI foundation models under the IndiaAI Mission. The models — EKA-1, Dhīr 2B and Dhīr 5B — are designed to support different AI applications, including mathematical reasoning, coding, multilingual communication, document automation and secure offline AI deployment.

The announcement was made on November 3, 2026, by the Ministry of Electronics and Information Technology (MeitY) at an event in New Delhi.

The models were developed by two Indian technology companies: Soket Labs Technology and Research Private Limited and ZenteiQ AiTech Innovations.

What makes this announcement interesting is that India is not focusing on just one type of AI system. Instead, the country is developing different models for different requirements, from advanced reasoning and coding to lightweight AI assistants that can operate in environments with limited internet connectivity.

In this detailed article, let's understand what these three AI models are, how they work, what makes them different, and why their launch matters for Indian students, developers, businesses and the future of India's AI industry.

1. What Happened? India Introduces Three Indigenous AI Models

Artificial intelligence is becoming an important part of everyday life. People use AI for writing, programming, education, research, business automation and content creation.

However, many popular AI services depend on cloud infrastructure. This means that users generally send their requests through the internet to servers that process the information and return an answer.

India wants to develop more AI capabilities within its own technology ecosystem.

As part of this effort, the IndiaAI Mission has supported the development of indigenous foundation models.

On November 3, the government announced three new models:

1. EKA-1: Developed by Soket Labs.

2. Dhīr 2B: Developed by ZenteiQ AiTech Innovations.

3. Dhīr 5B: Developed by ZenteiQ AiTech Innovations.

These models have different technical characteristics and intended applications.

EKA-1 focuses on mathematics, coding, reasoning and agentic applications. Dhīr 2B and Dhīr 5B focus on multilingual assistance, document-grounded automation and secure deployment in environments where internet connectivity may be limited.

The government has highlighted their potential use across education, government, industry, defence and other important sectors.

2. What Is EKA-1? India's New 26.3-Billion-Parameter AI Model

EKA-1 is an indigenous foundation model developed by Bengaluru-based Soket Labs Technology and Research.

One of its most interesting features is its scale.

According to the official announcement, EKA-1 contains approximately 26.3 billion parameters and was trained on around 7.5 trillion tokens.

But what do these technical terms actually mean?

Understanding parameters

Parameters are numerical values that an AI model learns during training. They help the model identify patterns, understand relationships between words and generate responses.

Generally, larger parameter counts indicate a model with more learned values. However, a higher parameter count does not automatically mean better performance.

Architecture, training quality, data, optimisation and evaluation also matter.

EKA-1 uses an open, sparse hybrid language-model architecture.

Training on 7.5 trillion tokens

Tokens are smaller units of text that AI systems process.

A token may represent a complete word, part of a word or punctuation.

Training on a large collection of tokens allows a model to learn language patterns, relationships and information from its training data.

EKA-1 was trained on approximately 7.5 trillion tokens, according to the government announcement.

Support for Indian languages

One of EKA-1's important features is its language coverage.

It supports all 22 scheduled Indian languages along with English.

This is particularly relevant for India because people communicate in different languages depending on their location, education and daily requirements.

An AI model with broad Indian-language support could help developers build applications for people who are not comfortable using English-only technology.

What can EKA-1 do?

The model is designed for several advanced applications:

Mathematical reasoning: It is intended to handle mathematical tasks and logical problem-solving.

Programming and coding: It is designed to assist with coding-related activities and software development.

Advanced reasoning: It targets tasks that require analysing information and working through multiple steps.

Agentic applications: It is designed for AI-agent use cases, where systems can work through sequences of tasks rather than simply generating a single response.

Soket has also highlighted applications related to India's banking, financial services and cybersecurity sectors.

This could create opportunities for specialised AI applications in industries where local requirements and domain knowledge are important.

3. What Are Dhīr 2B and Dhīr 5B?

The second part of the announcement introduces two models developed by ZenteiQ AiTech Innovations.

They are called Dhīr 2B and Dhīr 5B.

Unlike a large model designed primarily for complex reasoning, these models focus on multilingual assistance, document-based automation and deployment in environments where data privacy and connectivity are important.

The models are open-source and designed to support secure, on-premise deployment.

Dhīr 2B

Dhīr 2B is the smaller model in the family.

Its lightweight design makes it interesting for applications that do not require a very large AI system.

Potential applications include:

• Searching information inside documents.

• Retrieving information from organisational records.

• Automating repetitive document-related tasks.

• Supporting multilingual assistance.

• Building applications for low-connectivity environments.

Dhīr 5B

Dhīr 5B is the larger model in the Dhīr family.

It is designed for similar applications, including document-grounded automation, information retrieval and enterprise assistance.

The larger model size does not automatically guarantee better results in every application. Actual performance depends on the task, hardware, implementation and evaluation.

Language support

Both Dhīr models support English and 10 Indian languages.

This can help organisations develop applications that communicate with users in multiple languages.

For example, an educational institution could explore a multilingual assistant that helps students find information from approved academic documents.

A business could experiment with an internal assistant that searches company policies and retrieves relevant information.

These are potential applications rather than guaranteed capabilities in every deployment.

4. The Biggest Highlight: AI That Can Work Offline

One of the most interesting aspects of this announcement is the offline-capable deployment of Dhīr 2B and Dhīr 5B.

Most people are familiar with cloud-based AI applications.

When someone asks a question using a typical online AI service, the request is sent to remote computing infrastructure. The system processes the request and sends the result back.

On-device AI follows a different approach.

A compatible AI model can be installed and executed locally on a device or within an organisation's infrastructure.

This means that supported tasks can be performed without continuously depending on an external cloud service.

Why does offline AI matter?

Consider a school in a location with unreliable internet connectivity.

Students may still need help searching educational documents, finding information or using digital learning tools.

An appropriately configured local AI assistant could support these activities without requiring a continuous internet connection.

Now consider a hospital or government organisation that handles sensitive information.

Such organisations may prefer to keep certain information within their own infrastructure instead of sending every request to an external cloud service.

On-premise and offline-capable AI systems can provide another deployment option.

Four important advantages

1. Reduced internet dependency

Local AI applications can perform supported tasks even when internet connectivity is unavailable.

2. Greater control over information

Organisations can configure local deployments to keep data within their own systems.

3. Potential cost advantages

Local inference may reduce some recurring cloud-service costs, although hardware, electricity, maintenance and technical support still cost money.

4. More deployment flexibility

AI systems can be explored in educational institutions, industrial environments and other locations where cloud connectivity is difficult.

It is important to understand that offline deployment does not automatically guarantee complete security. Organisations still need appropriate access controls, encryption, updates and secure device management.

5. EKA-1 vs Dhīr 2B vs Dhīr 5B: What Is the Difference?

Although all three models were released under the IndiaAI Mission, they are designed for different requirements.

Here is a simplified comparison.

The comparison shows that the three models are not direct replacements for one another.

EKA-1 is designed for more advanced reasoning-oriented applications, while Dhīr focuses on lightweight multilingual assistance and local deployment.

The best model depends on what the developer or organisation wants to achieve.

6. How Can These AI Models Help Indian Students?

This announcement is particularly interesting for engineering students, computer science students and people learning artificial intelligence.

Students do not need to build a large language model from scratch to benefit from developments in AI.

They can learn how AI models are developed, deployed and integrated into real applications.

Here are some practical project ideas.

Project 1: Multilingual AI Study Assistant

Students could explore building an AI assistant that helps learners search educational materials in different Indian languages.

For example, a student could upload approved study documents and ask questions about a particular chapter.

The application could retrieve relevant information and provide a simplified explanation.

Skills involved:

• Python

• Natural Language Processing

• Language models

• Document processing

• Application development

Project 2: Offline Document Search System

A developer could create an application that searches through PDFs and other documents using a locally deployed AI model.

This could be useful for colleges, libraries and organisations that maintain large collections of documents.

Project 3: AI-Powered College Assistant

Imagine a college assistant that helps students find information about:

• Examination schedules.

• Attendance rules.

• Laboratory manuals.

• Academic regulations.

• Assignment deadlines.

• College notices.

The application could retrieve information from approved college documents.

This is an example of how AI can be integrated into a practical educational application.

Project 4: Indian Language Chatbot

Students could experiment with building a chatbot that supports Indian-language communication.

Such a project could be useful for educational support, customer assistance and information services.

These projects can help students build practical portfolios and demonstrate their understanding of AI technologies.

7. Opportunities for Indian Startups and Businesses

The launch is not only relevant to students.

Indian startups and established businesses could also explore specialised applications using indigenous AI models.

Banking and financial services

AI can support document processing, information retrieval, customer assistance and internal knowledge systems.

However, financial applications require careful testing, regulatory compliance and strong security controls.

Healthcare

Healthcare organisations could explore local AI assistants for searching approved medical documents and administrative records.

Any system used in clinical decision-making would require appropriate medical validation and human oversight.

Education

Educational institutions could explore multilingual learning assistants, document search systems and administrative automation.

Government services

Government organisations handle large amounts of documentation.

AI-powered retrieval and automation could help employees find relevant information and process repetitive tasks.

Manufacturing and industrial operations

Industrial organisations could explore AI assistants that search technical manuals, maintenance records and equipment documentation.

These are potential opportunities. The actual value will depend on reliability, implementation costs, security and the quality of the models.

8. Why Is the IndiaAI Mission Important?

The launch of these models is connected to a larger national initiative.

The IndiaAI Mission was approved by the Union Cabinet on March 7, 2024.

Its broader objective is to build an AI ecosystem that supports innovation, research, infrastructure and responsible AI development in India.

The mission includes seven important pillars.

1. IndiaAI Compute Capacity

Supports access to computing resources required for AI development and training.

2. IndiaAI Innovation Centre

Supports the development of indigenous AI models and foundation-model capabilities.

3. IndiaAI Datasets Platform

Supports access to datasets and related resources through AIKosh.

4. IndiaAI Application Development Initiative

Encourages AI applications that address important challenges.

5. IndiaAI FutureSkills

Focuses on developing AI skills and strengthening the talent pipeline.

6. IndiaAI Startup Financing

Supports the growth of startups working on AI technologies.

7. Safe and Trusted AI

Focuses on responsible AI development, safety and trust.

The development of EKA-1 and Dhīr demonstrates how public infrastructure and research support can contribute to domestic AI development.

According to the official announcement, the companies received access to subsidised national AI computing infrastructure and support for ancillary research expenses.

This type of support can be important because developing foundation models requires significant computing resources and specialised technical expertise.

9. Can These Models Compete With ChatGPT, Gemini and Claude?

This is one of the most common questions people may ask after hearing about India's new AI models.

The short answer is that there is not enough publicly verified comparative benchmark information in the launch announcement to conclude that these models outperform ChatGPT, Gemini or Claude.

Each model has a different design objective.

ChatGPT, Gemini and Claude are widely used AI services with different capabilities, products and ecosystems.

EKA-1 and Dhīr have been developed with specific Indian requirements and deployment opportunities in mind.

For example:

• EKA-1 focuses on mathematics, coding, reasoning and agentic use.

• Dhīr focuses on multilingual assistants and local deployment.

• Cloud-based AI services may offer broader general-purpose capabilities and integrated products.

A meaningful comparison would require independent testing across reasoning, coding, language understanding, accuracy, latency, safety and hardware requirements.

It would be incorrect to claim that the Indian models are already better than international AI systems without such evidence.

However, developing domestic alternatives is valuable even when the objective is not to outperform every global model.

India needs AI systems that address its own languages, industries, infrastructure and data requirements.

10. What Are the Challenges Ahead?

Despite the exciting possibilities, several challenges must be considered.

Model accuracy

AI systems can produce incorrect or misleading information.

Models used in important sectors must be tested carefully before deployment.

Hardware requirements

Running an AI model locally requires compatible hardware.

Performance can vary depending on memory, processing power and optimisation.

Language quality

Supporting a language does not automatically mean providing equally accurate results across every language and dialect.

Indian-language performance needs continuous evaluation.

Data privacy and security

Local deployment can improve control over information, but it does not eliminate security risks.

Organisations must protect devices, networks, model files and user data.

Real-world adoption

A model launch is only the beginning.

Developers and organisations need clear documentation, accessible tools, technical support and reliable deployment options to build useful applications.

Independent benchmarking

Transparent benchmarks can help users understand how these models perform against other available systems.

Independent testing will be important for building confidence among developers and businesses.

11. How Can You Explore These Models?

If you are a student, researcher or developer interested in experimenting with these models, the official IndiaAI ecosystem is a useful starting point.

You can explore the following resources:

• IndiaAI official website.

• AIKosh datasets and model resources.

• Official documentation from Soket Labs.

• Official documentation from ZenteiQ.

Before deploying any model, check its current availability, licence, hardware requirements and usage documentation.

For open-source models, read the licence carefully to understand what you can and cannot do with the model.

You can start with a small experiment, such as document retrieval or a multilingual question-answering application, before developing a larger project.

12. What Does This Mean for India's AI Future?

India has a large technology workforce, a growing startup ecosystem and a diverse linguistic environment.

These characteristics create opportunities to develop AI applications that address local needs.

For example, multilingual AI could help make digital services more accessible.

Offline AI could create new possibilities for organisations that cannot depend entirely on cloud connectivity.

Specialised reasoning models could support research, programming and business applications.

However, long-term success will depend on more than the announcement of new models.

India will need continued investment in research, computing infrastructure, high-quality datasets, skilled professionals, responsible development and independent evaluation.

It will also be important to create an ecosystem in which students, startups, researchers and businesses can experiment with these technologies.

The release of EKA-1, Dhīr 2B and Dhīr 5B is one step in that broader journey.

Frequently Asked Questions (FAQs)

1. Which three AI models has India launched?

India released EKA-1, Dhīr 2B and Dhīr 5B under the IndiaAI Mission on November 3, 2026.

2. Who developed EKA-1?

EKA-1 was developed by Soket Labs Technology and Research Private Limited.

3. Who developed Dhīr 2B and Dhīr 5B?

Both models were developed by ZenteiQ AiTech Innovations.

4. How many parameters does EKA-1 have?

EKA-1 has approximately 26.3 billion parameters and was trained on around 7.5 trillion tokens.

5. Which Indian languages does EKA-1 support?

It supports all 22 scheduled Indian languages along with English.

6. Can Dhīr models work offline?

Yes. Dhīr 2B and Dhīr 5B are designed for offline-capable and local deployment on compatible hardware.

7. Are Dhīr 2B and Dhīr 5B open-source?

Yes. They have been introduced as open-source models. Developers should check the official licence and documentation before using them.

8. Can engineering students use these models for projects?

Students can explore applications such as multilingual chatbots, document retrieval, educational assistants and local AI applications, subject to model availability and licence conditions.

9. Are these models better than ChatGPT?

There is insufficient independent comparative evidence to make that claim. Each model has different intended applications.

10. Where can users explore these models?

Users can explore the IndiaAI ecosystem and AIKosh for model-related resources and availability information.

Final Thoughts

India's launch of EKA-1, Dhīr 2B and Dhīr 5B represents an important development in the country's AI journey.

The announcement highlights three different directions for AI development: advanced reasoning, multilingual assistance and offline-capable deployment.

For students, it creates opportunities to learn and experiment. For developers, it opens possibilities for building applications around local AI models. For businesses, it provides additional options to explore specialised AI deployment.

The biggest opportunity is not simply using AI developed in India. It is building useful applications that solve real problems for Indian people and organisations.

India's next AI opportunity could come from the applications that developers, students and startups build using these technologies.