EKA-1, Dhīr 2B and Dhīr 5B: India’s Indigenous AI Foundation Models

Context

On 3 November 2026, the IndiaAI Mission under MeitY released indigenous AI foundation models EKA-1, Dhīr 2B and Dhīr 5B. Developed by Indian firms with IndiaAI support, the models strengthen domestic capability in multilingual, reasoning and locally deployable AI systems.

EKA-1: Large Multilingual Foundation Model

1. Developer: EKA-1 was developed by Soket Labs Technology and Research with support under the IndiaAI Mission.
2. Scale: It has 26.3 billion parameters and was trained on about 7.5 trillion tokens, giving it capacity for complex language and reasoning tasks.
3. Language capability: It supports all 22 Scheduled Indian languages and English, making it suitable for AI applications across India’s multilingual user base.
4. Architecture: Built as an open, sparse hybrid language model, it is designed for mathematics, coding, reasoning and agentic tasks, extending its use beyond basic conversational AI.

Dhīr 2B and Dhīr 5B: Lightweight Deployable Models

1. Developer: The open-source Dhīr models were developed by ZenteiQ AiTech Innovations.
2. Multilingual support: They work in English and 10 Indian languages, enabling local-language enterprise and citizen-service applications.
3. Flexible deployment: Their lightweight design allows operation on GPUs, TPUs and mobile devices, reducing dependence on large cloud infrastructure.
4. Secure local processing: They support on-premise and offline-capable deployment, making them useful where connectivity is weak or sensitive data should remain locally processed.

Applications

1. Government services: Document-grounded automation and knowledge retrieval can support multilingual citizen-assistance systems, improving access to public services.
2. Defence and sensitive sectors: Offline and on-premise operation reduces dependence on external cloud systems, making the models more suitable for confidential or low-connectivity environments.
3. Banking and cybersecurity: EKA-1 is designed for India-specific applications in sectors such as finance and cybersecurity, enabling more context-sensitive AI tools in regulated environments.
4. Education and enterprises: Multilingual assistants can support knowledge retrieval, document processing and workflow automation, widening practical AI use across institutions and businesses.

Significance for India’s AI Ecosystem

1. Strategic autonomy: Indigenous foundation models strengthen domestic control over model development, language adaptation and deployment choices, reducing excessive dependence on foreign proprietary systems.
2. Public AI infrastructure: India AI support, including access to subsidized compute and research resources, lowers the high cost of foundation-model development and can widen participation by Indian firms.
3. Diverse model ecosystem: Support for both a large model such as EKA-1 and lightweight models such as Dhīr shows a move towards different AI architectures for different Indian use cases rather than a single-model approach.
4. Innovation diffusion: Making indigenous models available to researchers and developers can encourage downstream applications, startups and experimentation across sectors.
5. Capability creation: Domestic foundation-model development helps India move beyond merely adopting global AI tools towards building core AI technologies and platforms of its own.

FAQs

Q1. Who developed EKA-1?
Soket Labs Technology and Research.

Q2. How large is EKA-1?
It has 26.3 billion parameters.

Q3. How many Indian languages does EKA-1 support?
All 22 Scheduled Indian languages, along with English.

Q4. Who developed Dhīr 2B and Dhīr 5B?
ZenteiQ AiTech Innovations.

Q5. What is the main advantage of the Dhīr models?
They support lightweight, multilingual, on-device and offline-capable AI deployment.