Soket Labs and ZenteiQ AiTech Innovations released indigenous language models under the IndiaAI Mission. The models were unveiled at an event in New Delhi with IT secretary S. Krishnan. They include Soket's EKA-1 and ZenteiQ's Dhīr 2B and Dhīr 5B.
Model details
EKA-1 has 26.3 billion parameters and was trained on roughly 7.5 trillion tokens. It supports all 22 scheduled Indian languages plus English. Soket designed it for mathematics, code generation, reasoning, and agentic applications.
Local deployment
ZenteiQ's Dhīr models are open-source and support English and 10 Indian languages. They can run on GPUs, TPUs, and mobile devices. This allows local use without cloud connectivity and supports offline document retrieval.
Sarvam's decision model
Separately, Sarvam AI launched Prism, an open-weight decision model. It is built on the Sarvam 105B model. The company reported an average score of 71.8 and a median response time of 44 milliseconds.
Both startups were selected under the IndiaAI Innovation Centre pillar. They received subsidized compute access and research cost support. Sarvam said Prism weights are available on Hugging Face and its own API.