What is Sarvam AI: Till now, America and China were considered to be dominant in the race of Artificial Intelligence, but India is also rapidly making its presence felt. of Bengaluru Startup Sarvam AI has developed its own AI The models have attracted the attention of the global tech world. The company claims that its models are performing better than many major global systems in tasks related to Indian languages like reading documents and producing voice etc. Its results, especially in document recognition (OCR) and text-to-speech technology, have attracted the attention of the tech world.
What is Universal AI?
Sarvam AI was founded in 2023 by Dr. Vivek Raghavan and Dr. Pratyush Kumar. Instead of building big and heavy cloud-based AI, the company aims to create smaller and faster models that serve India’s real needs. Internet speeds in India are not the same everywhere, and dozens of languages and dialects are used. Sarvam AI is working on solving this challenge i.e. AI that can work easily on mobile phones, call centers and in local languages. Earlier, Pratyush Kumar had launched AI4Bharat, an initiative to develop AI applications for Indian languages, and PadhAI, an affordable online education platform.
Drop 4/14: Introducing Sarvam Vision: a state-space based 3 billion parameter vision language model that is competitive with the best results in digitization in English, and defines a significantly higher bar for Indian languages. See the details in our blog:…
— Pratyush Kumar (@pratykumar) February 5, 2026
Why Bulbul V3 and Vision models are in discussion
Sarvam AI has recently launched two key technologies – Bulbul V3 (voice recognition system) and Vision (document reading OCR model). Currently, Bulbul V3 has more than 35 voice options for 11 Indian languages, and the company aims to increase this to 22 languages. Prateek Desai, founder of KissanAI, also praised it, saying that it is emerging as the most effective and affordable text-to-speech model for Indian languages. The Vision model showed better results than some common international models in reading documents written in Indian languages. Government papers, forms and records in India are often in local languages, so this technology can prove extremely useful.
Is it really better than Google and ChatGPT?
The tests cited by the company included blind listening tests and automatic error checks. Some media reports also mention votes from independent listeners and a large sample size. However, experts believe that such results cannot be considered as final conclusions. A true comparison of any new technology requires extensive and independent testing. So it is too early to say that one company has completely overtaken the global AI giants.
What benefit will the common people get?
If this technology is implemented on a large scale then many changes can be seen in India.
Call centers will be able to use cheaper and more natural voice assistants, government services will be more easily available in local languages, and digitizing documents will be faster.
Most AI models to date have been English-centric. But in a country like India, where linguistic diversity is very high, models based on local data and needs may be more effective.
Comments are closed.