India rolls out three sovereign AI models Sarvam AI, Gnani.ai, BharatGen to take on Big Tech

Representative image

• India unveiled three sovereign AI models at the AI Impact Summit
• Sarvam AI launched two large language models trained in India
• Gnani.ai and BharatGen launched voice and multilingual AI models

Did our AI summary help?

India on February 18 unveiled three sovereign artificial intelligence (AI) models at the India AI Impact Summit in New Delhi, as domestic startups and national initiatives stepped up efforts to build homegrown alternatives to global AI systems dominated by Big Tech.

The launches come against the backdrop of the IndiaAI Mission, which was approved by the Union Cabinet in March 2024 with an initial outlay of Rs 10,000 crore to build India’s sovereign AI capabilities, including a domestic foundational model, large-scale compute infrastructure, and AI applications for public use.

The mission gathered momentum in late 2024 and early 2025, with the government operationalising GPU subsidies and inviting startups to apply for compute support. Since then, over Rs 100 crore has been disbursed as subsidies for high-performance GPUs and over a dozen companies selected for the mission.

Bengaluru-based Sarvam AI, conversational AI firm Gnani.ai, and IIT-Bombay-led consortium’s sovereign AI initiative BharatGen announced a large language model (LLM) and voice models designed to be trained, deployed, and governed entirely within India.

“The developer energy I find in India is second to none. Recently, the work Sarvam has done developing local AI models, what you're talking about is actually happening. India is very well positioned,' Google CEO Sundar Pichai said at the India AI Impact Summit.

Two LLMs by Sarvam

First, the much-anticipated announcement by Sarvam AI, launching two large language models, a 30-billion-parameter model and a 105-billion-parameter model, both trained from scratch in India.

The company said the larger model outperforms global systems such as DeepSeek R1 and Google’s Gemini Flash on several benchmarks, while using a mixture-of-experts architecture to reduce inference costs.

The models are aimed at complex reasoning, programming, and agentic AI workloads, with Sarvam positioning efficiency as central to scaling AI for population-level use.

“Sovereignty matters much more in AI than building the biggest models,” said Sarvam co-founder Vivek Raghavan, at the Summit.

Gnani.ai text-to-speech model

Gnani.ai unveiled Vachana TTS, a text-to-speech model capable of cloning human voices across 12 Indian languages using under 10 seconds of reference audio.

The company said the model preserves voice characteristics such as tone, pitch, and speaking style, while allowing the same voice to operate across languages.

Built for low-bandwidth conditions and high-volume usage, Vachana TTS is targeted at government services, customer support systems, and large-scale enterprise deployments, with all data and models hosted within India.

BharatGen’s 17-bn parameter model

IIT Bombay-led consortium’s BharatGen launched its 17-billion-parameter multilingual foundational model, BharatGen Param2 17B MoE.

Param2 17B is a Mixture-of-Experts (MoE) model optimised for multiple Indic languages, with a focus on enabling AI adoption across governance, education, healthcare, agriculture and enterprise use cases.

BharatGen will be releasing the open-source model, documentation and post-training workflows through its Hugging Face repository. This will enable developers, startups and enterprises to build, fine-tune and deploy India-centric AI applications.

Having clocked in Rs 900 crores in funding from IndiaAI Mission, BharatGen is the biggest beneficiary so far of the government’s sovereign LLM initiative.