Sarvam is here, but will India’s startups use it?

Sarvam AI is trying to solve a problem with AI that none of the tech giants have addressed: India’s linguistic complexity.

In a country with 22 official languages where users frequently switch between them, localization becomes a necessity.

Sarvam co-founder and CEO Pratyush Kumar (first from left) at Razorpay FTX 2026./Photo credit: Razorpay.

With a sizable war chest and government backing, Sarvam carries symbolic weight. While Big Tech’s frontier models dominate the global market, Indian AI founders that Tech in Asia spoke with agree that there is both demand and need for localized models.

But the real test is whether its effort will result in real-world adoption. Local AI firms in other countries outside the US and China will also be watching closely. 

Sarvam did not respond to Tech in Asia’s queries.

The government-backed company, founded in Bengaluru by Vivek Raghavan and Pratyush Kumar in 2023, has recently introduced two large language models (LLMs) tailored for India’s linguistic diversity: Sarvam-30B and Sarvam-105B.

The Sarvam-30B model, with 30 billion parameters, is designed for use cases such as chatbots, customer support, and enterprise automation. The larger Sarvam-105B model, with 105 billion parameters, targets more advanced tasks like coding, analytics, and long-context reasoning.

Both models are built on a shared foundation, including a 262,000-token vocabulary designed to support 22 official Indian languages alongside English.

AI models use tokens to break down words so they can be processed by the system. Those with a higher number of tokens are more accurate, but they are also more expensive to use. For comparison, ChatGPT uses about 100,000 tokens.

Language wins, but is it enough?

Sarvam is being seen as a ChatGPT rival in India, offering similar conversational AI capabilities but tailored to the country’s languages and use cases.

At the launch of Sarvam’s models in February, Kumar had said they deliver “stronger performance than several larger competitors,” such as Google Gemini 2.5 Flash, on Indian language benchmarks.

Language accuracy in business can’t be overrated, according to founders that Tech in Asia spoke with.

Karl Chan, co-founder of Sourcy AI, points to practical use cases like supplier negotiations. In its system, which negotiates prices and minimum order quantities on behalf of customers, Sourcy sees better results when the model understands the supplier’s native language.

Moment of truth

Not one-size-fits-all