Sarvam AI Options In India: India’s sovereign AI journey is now not nearly huge world corporations opening workplaces or information facilities right here. At a time when it felt like India was nonetheless behind corporations like OpenAI and Google, a Bengaluru-based startup shocked everybody. The startup, known as Sarvam AI, has marked a brand new and essential step in India’s know-how story by creating a brand new home-grown AI fashions specifically made for Indian customers.
A Bengaluru-based startup, Sarvam AI, has launched two new AI fashions, Bulbul V3 and Sarvam Imaginative and prescient. What makes this improvement particular is that these fashions have outperformed well-liked instruments like ChatGPT and Google Gemini in duties comparable to studying and understanding textual content from photographs, a course of referred to as optical character recognition (OCR). This achievement highlights that India is now growing sturdy and superior AI options by itself.
Drop 6/14: @SarvamAI is proud to announce a landmark in India’s sovereign AI journey by way of strategic partnerships with the Governments of Odisha and Tamil Nadu. The purpose of those partnerships is to drive transformation by constructing at-scale compute, sovereign fashions, and the… pic.twitter.com/Scx9mK6CPw— Pratyush Kumar (@pratykumar) February 9, 2026
For a very long time, discussions round AI giant language fashions (LLMs) within the tech world have been dominated by the US and China. Regardless of India’s huge expertise pool and big AI market, the absence of a domestically developed AI mannequin had typically raised questions in regards to the nation’s place within the world AI race.
Sarvam AI beats ChatGPT, Gemini 3 Professional and DeepSeek
Sarvam AI has not too long ago gained consideration for delivering stronger outcomes than a number of main world AI fashions throughout key benchmarks. Its OCR answer, Sarvam Imaginative and prescient, secured the highest place on the olmOCR-Bench with an accuracy of 84.3 %, outperforming well-known instruments comparable to ChatGPT, Gemini 3 Professional, and DeepSeek OCR v2.
The mannequin additionally achieved a excessive rating of 93.28 % on OmniDocBench v1.5, displaying its skill to precisely deal with complicated web page layouts, technical tables, and mathematical equations. These are areas that always problem conventional OCR techniques. As well as, Sarvam AI has confirmed to be dependable for on a regular basis duties, together with scanned paperwork, kinds, and content material in a number of languages.
Bulbul V3 Options
The Bulbul V3, Sarvam AI’s text-to-speech mannequin, helps 35 completely different voices drawn from 22 official Indian languages, overlaying content material from as early because the 1800s to the current day. It’s designed to deal with various scan qualities and numerous forms of content material with accuracy. The collection additionally features a 3B-parameter state-space vision-language mannequin that may carry out superior visible understanding duties comparable to picture captioning, scene textual content recognition, chart evaluation, and the interpretation of complicated tables.
Drop 5/14: Introducing Bulbul V3, our newest text-to-speech mannequin. It raises the bar for the way human it sounds, whereas being tremendous strong.
In an impartial third-party human listening examine, Bulbul V3 delivers the best listener desire, and low error charges throughout use-cases… pic.twitter.com/w7HThWzuKe— Pratyush Kumar (@pratykumar) February 7, 2026
Sarvam Imaginative and prescient and Bulbul V3: The way it works
Sarvam Imaginative and prescient is designed as an India-first AI mannequin that understands the nation’s extensive cultural and language variety. It goals to construct a powerful AI basis developed inside India, making it a promising choice to be used in authorities tasks, public infrastructure, and the BFSI sector.
We additionally evaluated for the long-tail of language challenges comparable to talking numerics, technical content material, and named entities. Bulbul V3 constantly has the bottom error charges throughout languages. pic.twitter.com/1COxQU80J7— Pratyush Kumar (@pratykumar) February 7, 2026
In the meantime, Bulbul V3 is Sarvam AI’s flagship text-to-speech mannequin, constructed to deal with India’s wealthy and complicated language variety. It marks an enormous step ahead in creating pure, ready-to-use AI voices throughout a number of Indian languages.
One in every of its standout options is its easy language switching, permitting it to maneuver simply between languages like Tamil and English or Hindi and English with none disruption. Presently, the Bulbul V3 helps 11 Indian languages with over 35 voices, and Sarvam AI plans so as to add 22 extra Indian languages sooner or later.