Editorial illustration for IndiaAI Mission Releases 3 Indigenous AI Models by Soket, ZenteiQ
India Launches 3 Indigenous AI Models: EKA-1, Soket, ZenteiQ
IndiaAI Mission Releases 3 Indigenous AI Models by Soket, ZenteiQ
The IndiaAI Mission added three homegrown foundation models to its roster on November 3, unveiling them at an event in New Delhi attended by MeitY Secretary S Krishnan and other officials from the ministry and the mission. Soket AI Labs built EKA-1, a 26.3-billion-parameter sparse hybrid language model trained on roughly 7.5 trillion tokens. It handles all 22 scheduled Indian languages plus English and covers mathematics, coding, reasoning and agentic tasks.
ZenteiQ AiTech Innovations brought two models to the table, Dhīr 2B and Dhīr 5B, pitched as open-source multilingual assistants designed for offline and on-premise use. Government offices, defence units, factories and schools are the intended users, places where constant cloud access isn't always guaranteed or even wanted. Soket, for its part, pointed EKA-1 at a narrower target: India's banking, financial services and cybersecurity sectors, where data sensitivity and local context matter as much as raw model performance.
Both releases fit into the IndiaAI Mission's broader push to build AI infrastructure and models suited to Indian languages and institutions, rather than relying solely on foreign systems. Soket AI Labs CEO Abhishek Upperwal spoke about what went into EKA-1's design.
S Krishnan said the development of indigenous models would strengthen India's technological autonomy in artificial intelligence and create opportunities to improve productivity across government, industry, research and education.
Why this matters
For developers and founders building in Indian languages, EKA-1 and the Dhīr models are worth testing now, not just reading about. A 26.3-billion-parameter model trained on 7.5 trillion tokens with coverage of all 22 scheduled languages is a real claim, and the subsidised compute access S Krishnan mentioned lowers the cost of trying it out or fine-tuning for niche use cases that big labs skip. The Dhīr models' pitch, offline and on-premise deployment for government use, points at a market, public sector, healthcare, defense, that foreign closed models often can't touch because of data residency rules.
That's a genuine opening for teams building compliance-heavy products. But releasing a model isn't the same as proving it holds up against Llama or Qwen on real benchmarks across those 22 languages. Krishnan's framing about "technological autonomy" is a policy goal, not a technical one.
Watch for independent benchmark results and actual production deployments in the next few months. That's what will tell us whether this is a capability milestone or a subsidy-driven press release.
Related Reading
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source Box CEO Aaron Levie Says AI Will Drive "Biggest Cybersecurity Upgrade
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
From Archives Anthropic Releases FlashProve: GPU Proves Reply Came From Committed Weights
From Archives Soniox STT v6 adds text input, echo filtering; no v6 accuracy or latency
Common Questions Answered
What are the key specifications of EKA-1, the foundation model developed by Soket AI Labs?
EKA-1 is a 26.3-billion-parameter sparse hybrid language model trained on approximately 7.5 trillion tokens. It supports all 22 scheduled Indian languages plus English, and is capable of handling mathematics, coding, reasoning, and agentic tasks, making it a comprehensive solution for diverse AI applications across India.
How many indigenous AI models did the IndiaAI Mission release, and which companies developed them?
The IndiaAI Mission released three indigenous AI models on November 3. Soket AI Labs developed EKA-1, while ZenteiQ AiTech Innovations brought additional models to the roster, as unveiled at an event in New Delhi attended by MeitY Secretary S Krishnan and other ministry officials.
What benefits do the Dhīr models offer for government deployment according to the article?
The Dhīr models are specifically designed for offline and on-premise deployment, making them suitable for government use cases where data security and operational independence are critical concerns. This approach allows government agencies to deploy AI solutions without relying on cloud infrastructure or external connectivity.
Why does MeitY Secretary S Krishnan believe indigenous AI models are important for India?
According to S Krishnan, the development of indigenous models would strengthen India's technological autonomy in artificial intelligence and create opportunities to improve productivity across government, industry, research, and education sectors. This positions India to reduce dependence on foreign AI systems and build domestic capabilities.
What cost advantage do developers have when testing EKA-1 and other indigenous models?
The IndiaAI Mission provides subsidized compute access for developers and founders, which significantly lowers the cost of trying out or fine-tuning these models for niche use cases. This accessibility enables developers to test and customize models that large labs might otherwise overlook, democratizing AI development in India.
Further Reading
- Papers with Code - Latest NLP Research - Papers with Code
- Hugging Face Daily Papers - Hugging Face
- ArXiv CS.CL (Computation and Language) - ArXiv
