Krux
February 20, 2026
India Drops Three Sovereign AI Models for Local Languages
Published: February 20, 2026 at 12:56 AM
Updated: February 20, 2026 at 12:56 AM
100-word summary
India launched three homegrown AI systems at the India AI Impact Summit 2026 in New Delhi, targeting governance, education, healthcare, and enterprise needs. Sarvam AI rolled out two large language models (30 billion and 105 billion parameters) trained specifically for Indian languages, claiming performance on par with Gemma and Mistral at lower costs. BharatGen, an IIT-backed nonprofit, released Param2, a 17 billion parameter model supporting 22 languages with Nvidia's help. Gnani.ai introduced Vachana TTS, a voice cloning system for 12 Indian languages that needs under 10 seconds of audio. These sovereign models could reduce India's reliance on foreign AI infrastructure while ensuring data stays local, potentially reshaping how millions access...
What happened
India launched three homegrown AI systems at the India AI Impact Summit 2026 in New Delhi, targeting governance, education, healthcare, and enterprise needs. Sarvam AI rolled out two large language models (30 billion and 105 billion parameters) trained specifically for Indian languages, claiming performance on par with Gemma and Mistral at lower costs. BharatGen, an IIT-backed nonprofit, released Param2, a 17 billion parameter model supporting 22 languages with Nvidia's help. Gnani.ai introduced Vachana TTS, a voice cloning system for 12 Indian languages that needs under 10 seconds of audio.
Why it matters
These sovereign models could reduce India's reliance on foreign AI infrastructure while ensuring data stays local, potentially reshaping how millions access AI-powered services in their native tongues.