India AI models 2026 EKA-1 and Dhīr : What They Mean as Global AI Stocks Face a Sell-Off

India has taken another step towards developing its own artificial intelligence technology. On 3 November 2026, the IndiaAI Mission, under the Ministry of Electronics and Information Technology (MeitY), announced three indigenous AI foundation models: EKA-1, Dhīr 2B and Dhīr 5B. Developed by Indian AI companies Soket Labs and ZenteiQ, the models are designed for different requirements, ranging from advanced reasoning and multilingual AI to offline assistants that can operate without a continuous internet connection.

The announcement comes at a time when the global AI industry is facing a very different kind of development. On 5 November, technology and semiconductor stocks suffered a sharp sell-off as investors questioned the cost, profitability and future pace of AI infrastructure investment.

These developments are not directly connected, but together they highlight an important shift in the AI industry: countries and companies are racing to build useful AI systems, while investors are becoming more selective about how much they are willing to spend on the technology.

So, what are EKA-1 and Dhīr? How are they different from globally popular AI assistants? And could India’s indigenous models become useful in government, education, banking and business?

IndiaAI Mission Announces Three Indigenous AI Models

The Government of India’s announcement covers three models developed by two companies:

• EKA-1: Developed by Soket Labs Technology and Research Private Limited.
• Dhīr 2B: Developed by ZenteiQ AiTech Innovations.
• Dhīr 5B: Also developed by ZenteiQ AiTech Innovations.

The models were released under the Foundation Models pillar of the IndiaAI Mission. The initiative is intended to strengthen India’s domestic AI capabilities and encourage the development of technology suited to the country’s languages, industries and public-sector requirements.

According to the official government announcement, the selected companies received support through the IndiaAI Mission, including access to subsidised national AI computing infrastructure and assistance for research-related expenses.

This matters because training a foundation model requires substantial computing power, large datasets, specialised engineering and considerable financial resources. Public infrastructure support can help Indian research teams compete in an industry where access to advanced computing hardware is a major barrier.

Official announcement: Press Information Bureau – IndiaAI Mission releases EKA-1 and Dhīr models.

1. What Is EKA-1 AI Model?

EKA-1 is an indigenous foundation model developed by Bengaluru-based Soket Labs. It is designed for mathematical problem-solving, coding, reasoning and agentic applications, with support for all 22 scheduled Indian languages alongside English.

According to the Ministry of Electronics and Information Technology, EKA-1 has the following reported specifications:

• Model parameters: 26.3 billion.
• Training data: Approximately 7.5 trillion tokens.
• Language support: All 22 scheduled Indian languages and English.
• Architecture: Open, sparse hybrid language model.
• Designed capabilities: Mathematics, coding, reasoning and agentic use.
• Developer: Soket Labs Technology and Research Private Limited.

In simple terms, EKA-1 is intended to handle more than ordinary question-and-answer interactions. Its stated focus includes tasks that require structured reasoning, technical problem-solving and multi-step workflows.

Why Does EKA-1’s Indian Language Support Matter?

India has a large population that communicates in different languages. AI products that work primarily in English may not be equally accessible to people who prefer Hindi, Marathi, Tamil, Bengali, Telugu or other Indian languages.

Support for all 22 scheduled Indian languages could help developers build applications for a wider range of users. Potential applications include:

• Multilingual government information services.
• Education and learning assistants.
• Banking and financial service support.
• Technical documentation and knowledge retrieval.
• Indian-language business applications.
• Cybersecurity and enterprise workflows.

However, language support alone does not guarantee equal performance in every language. Real-world accuracy, response quality, safety and performance need to be evaluated through independent testing and practical use.

Who Developed EKA-1?

EKA-1 was developed by Soket AI, an Indian AI research company. The company’s Project EKA focuses on sovereign AI models, technical reasoning and multilingual capabilities.

The company’s broader research programme covers mathematics, code, reasoning and multilingual AI. Its work is aimed at building AI systems that can be deployed in environments where organisations need greater control over their infrastructure and data.

For further information, visit Soket’s Project EKA page.

2. What Are Dhīr 2B and Dhīr 5B AI Models?

While EKA-1 focuses on a broader foundation-model capability, Dhīr 2B and Dhīr 5B are designed around compact, multilingual AI deployment.

Developed by Indian company ZenteiQ, the Dhīr models are intended to work on local infrastructure, including laptops and edge devices. Their offline capability is particularly relevant for organisations that cannot depend on a continuous internet connection or prefer to keep sensitive information within their own systems.

The Government of India describes Dhīr 2B and Dhīr 5B as open-source multilingual assistants that can support secure, on-premise and offline-capable deployments.

Dhīr 2B: A Compact AI Model

Dhīr 2B is the smaller of the two released models. ZenteiQ’s model documentation lists it as a dense transformer model with approximately 2.03 billion parameters.

Its compact size is intended to make deployment more practical on laptops and edge devices, without requiring every interaction to be processed by a remote cloud service.

Dhīr 5B: A Mixture-of-Experts Model

Dhīr 5B uses a Mixture-of-Experts (MoE) architecture. ZenteiQ reports approximately 5.09 billion total parameters, with around 1.79 billion active per token.

In a Mixture-of-Experts model, different specialised components can be activated for different inputs rather than using every parameter for every token. This architecture can help manage computational requirements while supporting more complex tasks.

Key Features of Dhīr AI Models

• Language support: English and 10 Indian languages.
• Offline operation: Designed to operate without a continuous network connection.
• Deployment: Laptops, edge devices and on-premise infrastructure.
• Applications: Document-grounded automation, knowledge retrieval and enterprise assistants.
• Architecture: Dhīr 2B uses a dense transformer; Dhīr 5B uses a Mixture-of-Experts architecture.
• Licence: Apache 2.0, according to ZenteiQ’s model documentation.

The supported Indian languages listed by ZenteiQ include Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi and Odia, alongside English.

For detailed specifications and downloads, visit the official Dhīr model page.

3. EKA-1 vs Dhīr: What Is the Difference?

Although both are Indian foundation-model initiatives supported under the IndiaAI Mission, EKA-1 and Dhīr are not identical products. Their reported capabilities and intended deployment scenarios are different.

Note: Specifications are based on the government announcement and the developers’ published model documentation. Model size alone does not establish which model performs better. A fair comparison requires testing the models on the same benchmarks and hardware.

4. Why Are Offline AI Models Important for India?

Many popular AI services rely on cloud infrastructure. A user sends a prompt over the internet, the request is processed on remote servers, and the response is returned to the device.

This approach works well for many applications, but it may not be suitable for every organisation. Government departments, defence establishments, financial institutions, hospitals and industrial facilities can have strict data-security requirements.

Offline-capable models offer another deployment option. Instead of sending every request to an external service, an organisation can run an AI model on infrastructure under its own control.

Potential advantages include:

Data Control and Privacy

Organisations can process selected information locally, reducing the need to transmit sensitive documents to external cloud services. However, local deployment does not automatically make a system secure. Access controls, encryption, software updates and security audits remain essential.

Use in Low-Connectivity Locations

AI applications can be useful in locations where internet access is unreliable or unavailable. Offline deployment could support industrial sites, remote facilities and other environments with connectivity limitations.

Lower Dependence on External AI Services

Indian organisations may gain more choices when selecting AI infrastructure. They can evaluate locally developed models alongside international services and decide which deployment approach meets their technical, financial and regulatory requirements.

Government and Enterprise Applications

The IndiaAI Mission announcement identifies defence, government, industrial and educational sectors as potential application areas for Dhīr. EKA-1’s developer has also highlighted banking, financial services and cybersecurity as areas of interest.

These are potential use cases, not proof that the models have already been widely adopted or deployed across these sectors.

5. What Is the IndiaAI Mission and Why Is It Supporting These Models?

The IndiaAI Mission is a Government of India initiative approved by the Union Cabinet on 7 March 2024. It operates under the Ministry of Electronics and Information Technology and aims to build a stronger domestic artificial intelligence ecosystem.

The mission is based on the principle of Making AI in India and Making AI Work for India.

It is organised around seven pillars:

1. IndiaAI Compute Capacity: Expanding access to computing infrastructure for AI development.
2. IndiaAI Innovation Centre: Supporting the development of indigenous AI models and research.
3. IndiaAI Datasets Platform: Improving access to datasets and resources through AIKosh.
4. IndiaAI Application Development Initiative: Encouraging AI applications for important sectors and public needs.
5. IndiaAI FutureSkills: Developing AI skills and talent.
6. IndiaAI Startup Financing: Supporting startups working in the AI ecosystem.
7. Safe and Trusted AI: Promoting responsible, reliable and secure AI development.

Support for EKA-1 and Dhīr comes under the IndiaAI Innovation Centre’s foundation-model initiative. Access to subsidised computing infrastructure is particularly important because training large AI models can be expensive and resource-intensive.

Another important resource is AIKosh, the IndiaAI datasets platform. It provides access to datasets, models, toolkits and other resources that can help researchers and developers build AI applications.

For background, read the government’s explanation of the IndiaAI Mission and AIKosh.

6. Meanwhile, Global AI Stocks Face a Sharp Sell-Off

While India announced new indigenous AI models, global financial markets were dealing with a sharp decline in technology and semiconductor stocks.

On 5 November 2026, a major sell-off hit US technology shares. Reuters reported that the Philadelphia Semiconductor Index fell 12.2% during the session, its largest one-day decline since April 2025. The Nasdaq Composite fell 5.5%, while the S&P 500 declined 4.4%.

Major AI-linked companies were also affected. Nvidia shares fell 9.3%, while Qualcomm dropped after its smartphone chip sales missed market expectations.

The decline reflected several investor concerns:

• AI infrastructure spending: Investors are questioning whether the enormous investments in data centres, chips and computing infrastructure will generate sufficient returns.
• Data-centre expansion: Higher borrowing costs and political opposition to some data-centre projects have raised concerns about the pace of future construction.
• Semiconductor demand: Weak results from some chip-related businesses added pressure to an already cautious market.
• Valuations: After a major rally in AI-related stocks, investors are reassessing whether current valuations reflect realistic future earnings.
• Policy uncertainty: The US midterm election results introduced additional uncertainty about the future regulatory environment for AI infrastructure.

The sell-off also spread into Asian markets. On 6 November, several major Asian technology and semiconductor markets came under pressure as investors reacted to the previous session’s losses in the United States.

Sources: Reuters – Nasdaq correction and AI spending concerns and Associated Press – Asian markets react to the AI stock sell-off.

7. Does the Global AI Stock Sell-Off Mean the AI Boom Is Over?

Not necessarily.

A fall in AI-related stock prices does not automatically mean that demand for artificial intelligence is disappearing. Financial markets price expectations about future earnings, growth, costs and risks. When investors believe that companies are spending too much or may take longer to generate returns, stock prices can fall even while the underlying technology continues to develop.

The current sell-off highlights a distinction between two questions:

• Is AI technology becoming more capable and useful? This depends on research progress, model performance, adoption and real-world applications.
• Are AI-related companies worth their current market valuations? This depends on revenue, profitability, spending, competition and investor expectations.

These questions are related, but they are not the same. AI adoption can continue growing even if some AI stocks experience a correction.

For India, the launch of EKA-1 and Dhīr is part of a longer-term effort to develop domestic capabilities rather than simply depend on international AI platforms.

8. Can India’s Indigenous AI Models Compete With Global AI Companies?

India’s new models represent progress, but it is too early to declare them direct competitors to the world’s largest AI systems.

Companies such as OpenAI, Google, Anthropic and Meta operate at a different scale, with extensive research teams, computing infrastructure, products and global distribution networks.

However, competing does not always mean building a model that is better at every task. India can also create value by developing models that are particularly useful for Indian languages, local industries, regulated environments and organisations that require on-premise or offline deployment.

For example, a smaller model that can run locally and retrieve information from an organisation’s documents may be more practical for a particular business than a much larger cloud-based model.

India’s opportunity may therefore lie in building specialised, efficient and locally relevant AI systems, rather than trying to reproduce every capability offered by global technology companies.

What Needs to Happen Next?

The long-term success of these models will depend on several factors:

• Independent performance testing: Results across Indian languages, coding, mathematics and reasoning need to be evaluated using transparent benchmarks.
• Real-world adoption: Government departments, educational institutions and businesses must find practical reasons to deploy the models.
• Developer ecosystem: Researchers and software developers need access to model weights, documentation, tools and reliable deployment options.
• Safety and accuracy: Models used in banking, defence, education or public services require strong safeguards and evaluation.
• Continued investment: Training and maintaining competitive models requires sustained funding, computing infrastructure and technical talent.

India has taken another step in building its own AI ecosystem. Whether these models become widely used products will depend on what happens after their announcement.

Conclusion: India’s AI Ambition Is Moving Beyond Adoption

The launch of EKA-1, Dhīr 2B and Dhīr 5B marks an important development in India’s effort to build indigenous artificial intelligence capabilities.

EKA-1 brings a focus on mathematical reasoning, coding, agentic applications and broad Indian-language support. Dhīr takes a different approach, with compact models designed for offline and on-premise deployment.

At the same time, the global AI-stock sell-off shows that enthusiasm for artificial intelligence does not remove the need for commercial discipline. Investors are increasingly looking for evidence that large AI investments can translate into sustainable returns.

For India, the next challenge is not simply to announce more AI models. It is to make them reliable, accessible, secure and useful in real-world applications.

The real measure of success will be adoption: whether Indian businesses, institutions, developers and ordinary users find practical value in AI built in India.

Frequently Asked Questions (FAQs)

1. What are the new AI models launched by India in November 2026?

The IndiaAI Mission announced three indigenous AI models on 3 November 2026: EKA-1, developed by Soket Labs, and Dhīr 2B and Dhīr 5B, developed by ZenteiQ.

2. Who developed the EKA-1 AI model?

EKA-1 was developed by Soket Labs Technology and Research Private Limited, an Indian AI research company. The model has 26.3 billion parameters and was trained on approximately 7.5 trillion tokens, according to the government announcement.

3. What is the difference between EKA-1 and Dhīr AI?

EKA-1 is a 26.3-billion-parameter foundation model designed for mathematics, coding, reasoning and agentic use. Dhīr 2B and Dhīr 5B are compact multilingual models designed for local, offline and on-premise deployment.

4. Can Dhīr AI work without an internet connection?

Yes. ZenteiQ describes Dhīr 2B and Dhīr 5B as models designed to run offline on laptops and edge devices. The exact experience will depend on the hardware and software setup.

5. Which Indian languages are supported by Dhīr AI?

Dhīr supports English and 10 Indian languages: Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi and Odia.

6. How many Indian languages does EKA-1 support?

According to the Ministry of Electronics and Information Technology, EKA-1 supports all 22 scheduled Indian languages alongside English.

7. What is the IndiaAI Mission?

The IndiaAI Mission is a Government of India initiative approved in March 2024 to strengthen India’s artificial intelligence ecosystem through computing infrastructure, indigenous model development, datasets, skills, startup support and safe AI initiatives.

8. Where can users access Dhīr AI models?

ZenteiQ provides model information and download links through its official Dhīr model page. The government also identifies AIKosh as a platform for accessing AI resources.

9. Why did global AI stocks fall in November 2026?

AI-related stocks fell sharply on 5 November 2026 amid concerns about semiconductor earnings, data-centre spending, borrowing costs, infrastructure expansion and the time required for AI investments to generate returns.

10. Does the AI stock market sell-off mean artificial intelligence is losing demand?

Not necessarily. A stock-market correction reflects changing investor expectations and valuations. It does not, by itself, prove that demand for AI technology or adoption is declining.

Editorial note: This article is based on the Government of India’s announcement and publicly available developer documentation as of 6 November 2026. Model specifications and capabilities may change as developers publish further technical reports, evaluations and updates.