Amid a push for sovereign AI, gen AI start ups Sarvam and Gnani.AI have launched their own advanced models. The models released today, are well versed in diverse Indian languages, claim the companies.
The model supports 32,000 tokens and is available in 22 Indian languages. While Sarvam-105B supports a context length of 128,000 tokens, it means the model can read and remember a very large amount of text at once, roughly the length of a long book chapter or even several long documents together.
“Today we show we can bring our own AI to a billion Indians,” said Pratyush Kumar, co-founder, Sravam at the event.
3 February 2026
Get the latest issue of Outlook Business
At the launch, the company compared Sarvam-30B with other well-known AI models like Gemma 27B, Mistral-32-24B, OLMo 31.32B, Nemotron-30B, Qwen-30B and GPT-OSS-20B. These models were tested on tasks that check how well they solve maths problems and write correct computer code.
Several generative AI start-ups are using the IndiaAI Mission as a platform to launch their own models. For instance, Gnani.AI has introduced Vachana STT, an enterprise-grade speech-to-text model built for Indian languages. The company said the model has been trained on more than 1 million hours of real-world voice data across 1,056 different domains, making it suitable for a wide range of industries. Vachana STT is also the first model to be launched under Inya VoiceOS.
For context, Inya VoiceOS is a five-billion-parameter voice-to-voice foundational AI model.
Speaking to Outlook Business, at the sidelines of the summit, Ananth Nagaraj, co-founder and CTO of Gnani.ai, said, “The first version is a 5-billion parameter model that currently supports six languages. We plan to scale it to 12 languages with a 14-billion parameter version, and eventually to a 70-billion parameter model within the next year.”
Further, genAI start-up BharatGen is also supposed to launch its 17-billion-parameter multilingual AI model at the summit. As per the company, the model, called Param2 17B, has been built from the ground up by BharatGen and is designed as a Mixture of Experts (MoE) system trained extensively on Indian data, the consortium said.
All these start-ups are a part of the IndiaAI mission. This comes at a time when the the government has been increasing its push for sovereign AI. “Sovereignty matters much more in AI than building the biggest models,” said Sarvam co-founder, Vivek Raghavan, at the event.
The entire IndiaAI summit has been several dialogues among leaders of the global south to make AI inclusive and to prevent the domination of power in the hands of the global north.
However, this comes at a time when the country as a whole underspends on R&D, hovering around 0.6%–0.7% of GDP. To this, Nagaraj says, “As for R&D investment, yes, we need to invest more as a country. We have not done enough in the past. If India wants to become a true product nation, higher investment in research and development is essential. That will be a key factor in strengthening our AI ecosystem.”
