Soket Labs and ZenteiQ AiTech Innovations have introduced new indigenously developed large language models (LLMs) designed to support India’s diverse languages and specialized use cases under the government-backed IndiaAI Mission.
- Soket’s EKA-1 supports 22 Indian languages and focuses on mathematics and coding.
- ZenteiQ’s Dhīr models are optimized for on-device deployment without cloud reliance.
- IndiaAI Mission subsidized AI compute resources to enable the startups’ developments.
What happened
In New Delhi, two Indian AI startups, Soket Labs and ZenteiQ AiTech Innovations, launched new indigenous large language models developed under the IndiaAI Mission. The event was attended by IT secretary S Krishnan, highlighting governmental support for domestic AI innovation. Soket Labs unveiled its EKA-1, a 26.3 billion parameter open, sparse hybrid language model trained on 7.5 trillion tokens, tailored to support a range of functions from mathematical problem solving to code generation and agentic applications.
Simultaneously, ZenteiQ introduced its Dhīr 2B and Dhīr 5B models that cater to multilingual usage, supporting English plus 10 Indian languages. These models are designed to operate in resource-constrained environments with offline and on-device capabilities, enhanced by a lightweight architecture allowing deployment on GPUs, TPUs, and mobile devices without relying on cloud connectivity.
Why it matters
The launch of these indigenous models represents a milestone in India’s strategic push to bolster its technological self-reliance in artificial intelligence. By developing foundation models optimized for the country's linguistic diversity and industry needs—especially banking, finance, cybersecurity, government, and education—these AI advancements aim to enable tailored digital solutions that reflect India's unique ecosystem.
The models’ design considerations, such as multilingual support and offline usability, address critical local challenges like connectivity limitations and data privacy. This fosters broader AI adoption and empowers sectors that require secure, localized AI applications, ultimately contributing to productivity improvements across government and industry sectors.
What to watch next
Stakeholders will be closely observing the deployment timelines, performance benchmarks, pricing, and real-world application outcomes of Soket's and ZenteiQ's models, especially in sectors such as banking and education where customized AI tools can have high impact. Additionally, the possibility of integration with other domestic AI initiatives like Sarvam AI’s decision model Prism will be important to watch as India builds a comprehensive AI ecosystem.
Furthermore, the role of government subsidies and infrastructure support through the IndiaAI Mission will remain critical to the sustained development and scaling of native AI technologies. Monitoring how these indigenous models evolve and compete with global AI frameworks will provide insight into India’s growing position in the competitive enterprise AI market projected to reach $71 billion by 2030.