India Launches Three Homegrown AI Models For Banking, Cybersecurity And Offline Use

India has released three indigenous artificial intelligence foundation models designed to support Indian languages, cybersecurity, financial services and sensitive government operations, marking a fresh development under the Centre’s IndiaAI Mission.

The models, EKA-1 developed by Soket Labs and Dhīr 2B and Dhīr 5B developed by ZenteiQ, were unveiled in New Delhi on November 3, 2026, in the presence of Ministry of Electronics and Information Technology Secretary S. Krishnan and senior IndiaAI officials.

The launch brings together two different approaches to domestic AI development. While EKA-1 is a larger model intended for complex reasoning and specialised enterprise work, the Dhīr models are designed for smaller computing environments, including mobile devices that can operate without constant internet connectivity.

Both companies received support through the IndiaAI Mission, which is seeking to develop domestic AI capabilities and reduce dependence on technologies built and controlled overseas.

EKA-1 Targets Banking, Cybersecurity And Complex Reasoning

Soket Labs has developed EKA-1, an AI foundation model containing 26.3 billion parameters and trained on approximately 7.5 trillion tokens.

Parameters are the internal values an AI model learns during training, while tokens are smaller units of text and other information used to teach it patterns and relationships.

EKA-1 supports all 22 languages listed in the Eighth Schedule of the Indian Constitution, alongside English. It is designed to perform tasks involving mathematics, computer programming, reasoning and AI-assisted operations.

The model uses a sparse hybrid architecture, an approach intended to manage computing resources while handling different types of tasks.

Soket Labs Chief Executive Officer Abhishek Upperwal said the system had been developed with India’s banking, financial services and cybersecurity requirements in mind.

These sectors frequently handle confidential information and require technology that can operate within controlled environments.

Banks, for example, could explore specialised AI systems for analysing documents, assisting with compliance tasks or supporting internal cybersecurity processes.

However, the launch announcement does not establish that EKA-1 has already been adopted by banks or deployed in operational cyber-defence systems.

Its performance in such environments will depend on technical testing, security assessments and evaluations against existing alternatives.

Dhīr Models Designed To Work Without Internet Access

ZenteiQ has introduced two smaller AI systems, Dhīr 2B and Dhīr 5B, designed to support English and 10 Indian languages.

Unlike large AI services that rely heavily on remote cloud infrastructure, the models are intended to operate on local computing systems, including compatible mobile phones.

They can also be deployed within an organisation’s own infrastructure, commonly called on-premise deployment.

This approach could help organisations process information without routinely transmitting sensitive material to external AI servers.

ZenteiQ said its models can support document-based automation, information retrieval and multilingual assistance across defence, government, industrial and educational settings.

Their offline capabilities could also be useful in areas where internet connectivity is limited or unreliable.

The company has described the models as open-source, potentially allowing developers and researchers to inspect, adapt and use them under the applicable licensing terms.

Their smaller size may make deployment less expensive than operating much larger AI systems, although actual requirements will depend on hardware, configuration and workload.

Local processing can reduce some data-exposure risks, but it does not automatically make an AI system secure. Organisations must still protect devices, restrict access and test models for vulnerabilities and inaccurate outputs.

IndiaAI Mission Moves Towards Domestic Model Development

The launches form part of the IndiaAI Mission’s Innovation Centre initiative, which supports domestic developers building foundation models and large language models.

The Union Cabinet approved the IndiaAI Mission on March 7, 2024, as part of a broader effort to expand computing infrastructure, AI research, startup development and access to datasets.

Under the programme, selected developers receive support that includes subsidised computing resources and assistance with research expenses.

Such infrastructure is important because training advanced AI models requires substantial computing power, specialised expertise and large quantities of data.

The government also operates AIKosh, a platform intended to make AI-related resources accessible to researchers, students, startups and other users.

MeitY Secretary S. Krishnan said indigenous models could strengthen India’s technological and strategic autonomy while supporting productivity across government, industry, research and education.

The announcement also reflects growing interest in AI systems that can be adapted for local languages and specialised requirements rather than relying entirely on general-purpose international models.

For India, the significance extends beyond producing AI software domestically. Institutions handling sensitive financial, industrial or government data may benefit from having more choices over where AI systems operate and how information is processed.

Nevertheless, the launch represents an initial technology milestone rather than proof of large-scale commercial success.

The government has not supplied comprehensive independent benchmark results demonstrating how the three models compare with established international alternatives, or detailed figures on their current deployment.

Their wider impact will depend on practical performance, accessibility, developer adoption and whether organisations can integrate them securely into existing systems.

What This Means For You:

Indian-language AI could make digital tools easier to use, while offline models may give organisations greater control over sensitive data. Users and businesses should assess accuracy, security and privacy safeguards before relying on such systems for important decisions.