Last Updated:
Sarvam AI is made in India, focuses on languages in the country but has scored the global heavyweights.
This made in India AI model is making some tall claims. (Photo: AI generated)
India is crying out for locally built AI models to give the tech behemoths like OpenAI and Google a run for their money. And now, one company claims it has not only done that but also outperformed the global competition. Sarvam AI takes the laurel for this big feat, at least going by what the company’s CEO, Pratyush Kumar has said with a test chart to make his point.
His post says Sarvam Vision, the company’s AI chatbot rivaling ChatGPT and Gemini has outperformed the big AI models in the overall performance of certain aspects. The locally made AI model scored over 93 percent in the OmniDocBench v1.5 benchmark test, as the post here claims.
Sarvam Vision achieves state-of-the-art accuracy of 84.3% on the olmOCR-Bench (English only subset) outperforming frontier models like Gemini 3 Pro and recent OCR models like DeepSeek OCR 2. pic.twitter.com/NsvN15hNqa— Pratyush Kumar (@pratykumar) February 5, 2026
India’s Sarvam AI Better Than ChatGPT?
The benchmark is done to test the ability of AI chatbots and see how they perform while identifying complex fonts, handwriting and other versions of textual data. The 93.28 percent score is far higher than what ChatGPT and Gemini got in these tests.
This certainly is a big feat for an AI model that is not built with the resources available to OpenAI and Google, still manages to deliver a product that can easily handle some important tasks better than the biggest and faster models operating in the market.
So what is the reality about Sarvam Vision and how does it manage to beat the biggies? The context around these tests is very important to understand. It seems the made in India AI model has been trained on Indic scripts, which may be aiding its better delivery and responses.
While Google and OpenAI are good, their focus is more widespread, and also having over 3 trillion parameters mean these AI models are equipped to handle more than a few tasks. Sarvam AI is based on 3 billion parameters, built to handle visual language models and betting higher on Indian languages. It can also understand from visuals and describe what it can see.
The conditions are specific to particular regions and that’s fine but the promise shown by Sarvam AI along with its text-to-speech model called Bulbul V3 suggests the Indian startups are evolving and we could see more players enter the battleground in the near future.
Delhi, India, India
February 11, 2026, 09:41 IST
Read More
