Artificial intelligence is becoming an important part of everyday life. From chatbots and translation tools to coding assistants and business automation, AI is changing how people work and access information.
Until recently, many widely used AI models were developed by major international technology companies. India is now investing in its own foundation models to strengthen domestic AI capabilities and improve support for Indian languages.
On 3 November 2026, the IndiaAI Mission announced three indigenous AI foundation models: EKA-1, Dhīr 2B and Dhīr 5B.
These models were developed by two Indian technology companies, Soket Labs and ZenteiQ.
What Is the IndiaAI Mission?
The IndiaAI Mission is an initiative of India's Ministry of Electronics and Information Technology (MeitY).
Its objective is to strengthen India's AI ecosystem through computing infrastructure, research, startup support, datasets, skills development, and responsible AI initiatives.
Developing homegrown foundation models is one part of this wider strategy.
1. EKA-1 by Soket Labs
EKA-1 is a foundation model developed by Soket Labs Technology and Research.
According to the official government announcement, it has:
- 26.3 billion parameters
- Approximately 7.5 trillion training tokens
- Support for all 22 scheduled Indian languages and English
- Intended capabilities in mathematics, coding, reasoning, and agentic tasks
The company describes EKA-1 as an open, sparse hybrid language model.
Its stated areas of focus include banking, financial services, and cybersecurity.
However, these specifications and intended uses should not be confused with independent proof of performance. Public benchmark results and model-weight availability should be checked before drawing conclusions about its capabilities.
2. Dhīr 2B by ZenteiQ
Dhīr 2B is a compact foundation model developed by ZenteiQ.
According to the developer's documentation, it has approximately 2.03 billion parameters and supports English and ten Indian languages:
- Hindi
- Bengali
- Tamil
- Telugu
- Marathi
- Gujarati
- Kannada
- Malayalam
- Punjabi
- Odia
The model is designed for local deployment, including use on laptops and edge devices.
One important advantage of local AI is that certain tasks can be performed without continuously sending information to an external cloud service.
This can be useful in environments where privacy, connectivity, or control over data is important.
3. Dhīr 5B by ZenteiQ
Dhīr 5B uses a mixture-of-experts architecture.
It contains approximately 5.09 billion total parameters, with around 1.79 billion active per token according to ZenteiQ.
A mixture-of-experts model uses different groups of specialised neural-network components for different parts of a task. A routing mechanism selects which experts are used for each token.
This can reduce the computation required for each token compared with activating every parameter.
Dhīr 5B is also designed for offline and on-device deployment.
What Does Offline AI Mean?
Many popular AI services require an internet connection because their models run on remote servers.
Offline AI works differently. A compatible model can run directly on a laptop, mobile device, or local computer.
Potential advantages include:
- Reduced dependence on internet connectivity
- Greater control over locally stored information
- Possible use in remote locations
- More control over deployment and customisation
However, offline operation does not automatically guarantee privacy or security. Device security, model configuration, software updates, and data handling still matter.
Why Is Tamil Language Support Important?
India has many languages, and people do not always communicate using English.
AI models that support Indian languages can potentially help with information retrieval, translation, document processing, education, and business services.
Dhīr explicitly lists Tamil among its supported languages. EKA-1 is reported by the government to support all 22 scheduled Indian languages.
The actual quality of responses in each language must still be tested. Language support does not necessarily mean equal performance across all languages or tasks.
Can These Models Compete With ChatGPT and Gemini?
It is too early to make that conclusion.
Parameter counts, training-token counts, and language coverage provide useful technical information, but they do not independently establish how well a model performs.
A fair comparison requires consistent benchmarks covering areas such as:
- Tamil and other Indian-language understanding
- Mathematics and reasoning
- Coding
- Factual accuracy
- Document processing
- Safety and reliability
- Real-world task performance
Independent testing will be important in understanding where these Indian models are strongest.
Are the Models Available to Download?
Availability is an important issue.
An independent report published on 4 November 2026 found that EKA-1's AIKosh listing did not yet contain downloadable model weights. The same report noted that access to Dhīr downloads required approval at that time.
Users should check the current AIKosh listings and developer websites rather than assuming all three models are immediately available to everyone.
Conclusion
The launch of EKA-1, Dhīr 2B and Dhīr 5B represents another step in India's effort to build its own AI technology.
The models have different goals: EKA-1 targets broader multilingual and technical tasks, while Dhīr focuses on compact multilingual AI that can operate in local and offline environments.
Their long-term value will depend on real-world performance, accessibility, reliability, and continued development.
India's AI future will not be decided by model launches alone. It will also depend on the researchers, developers, students, businesses, and users who build useful applications with these technologies.
For more Tamil explanations of artificial intelligence and emerging technology, follow Neural AI Mind.
Sources
- Press Information Bureau / MeitY — IndiaAI Mission foundation model announcement
- ZenteiQ — Dhīr 2B and Dhīr 5B specifications
- ZenteiQ Research — Training small models from scratch
- Transclew — Independent investigation of model download availability
- The Economic Times — IndiaAI Mission model launch coverage
