If you have been following the recent AI space, you might have heard of ‘Sarvam AI’. Recently, this name has started appearing in many AI benchmark discussions, quite consistently, alongside big names like ‘ChatGPT’ and ‘Gemini’.
There have also been claims that Sarvam AI India is surpassing ChatGPT and Gemini in some specific real world benchmarks like OCR accuracy, Indian language speech and context heavy use cases.
Now imagine, almost everyone, from a techie to a non- techie has been relying on these LLM’s for almost all types of tasks. Then emerges an Indian startup that starts outperforming these giant LLM’s in some specific use cases. This is the part that made people start noticing it.
So what’s actually happening here? And why are international researchers and industry observers suddenly mentioning an Indian AI startup, simultaneously with the largest AI platforms in the world?
Let’s break it down.
What Is Sarvam AI?
To put it plainly, Sarvam AI is an Indian startup that is building AI models which are specific to Indian use cases. The goal for Sarvam AI has been very clear since day 1 – build a system that actually understands how Indian languages are actually spoken, written and mixed in real life.
This is a game changer for an AI Model, because Sarvam AI is dealing beyond the Hindi or English language. It has the capacity to deal with Tamil, Telugu, Kannada, Bengali, Marathi – basically languages that have very different scripts, structures and accents that the global models usually struggle to understand.
Now, this is a standing out feature amongst the huge AI models, simply because Sarvam AI India narrowed down its focus to solving a simple (but huge) issue, instead of covering everything at once. And the best part is that these results started showing up, not just in demos, but in places like deployments, practical performances, benchmarks etc. making it stand equally amongst ChatGPT and Gemini.
Sarvam AI vs ChatGPT vs Google Gemini: Where It’s Winning
Comparing an Indian startup AI with ChatGPT and Google Gemini, is where people are really starting to pay attention globally.
As reported by India Today, Sarvam AI’s co-founder, Pratyush Kumar shared some important performance details from the company’s in house models like Sarvam Vision and Bulbul V3, that have outperformed LLM’s, especially in context of Indian use cases.
1. OCR Performance That Actually Works in India
Sarvam Vision, the vision model developed by Sarvam AI, has reportedly surpassed Google Gemini 3 Pro, DeepSeek OCR v2 and even ChatGPT, on the olmOCR-Bench, recording an accuracy score of 84.3%.
But the Indian model performance extends beyond a single benchmark. Sarvam AI recorded an overall score of 93.28% on the OmniDocBench v1.5, which tests how the AI systems read and interpret the real world documents. The strongest results came from complicated layouts, dense tables and mathematical content – areas where OCR systems tend to struggle because of discrepant formatting and visual mess.
This is great news for the banking sector, government departments and enterprise operations because in such environments, the OCR accuracy directly affects the reliability of automation systems.
2. Indian Language Speech
Sarvam’s Bulbul V3, which is a text to speech model, has been gaining a lot of traction for its natural handling of Indian languages in real world use. The system currently supports more than 35 voices across 11 Indian languages, and they have plans to expand coverage to 22 languages.
Bulbul V3 stresses on the expressive and production ready speech. The voices are designed in a way that they match with how the Indian languages are actually spoken – which includes variations in tone and cadence that the global TTS systems miss out on, very often. This feature alone sets thisIndian model apart from the other AI models.
According to India Today, Pratik Desai, who is the founder of KissanAI, praised BulBul and referred to it as their ‘go to TTS model’ for Indic use cases. He added that the quality has consistently improved with each release. He also pointed out that the cost structure for global alternatives like ElevenLabs never made sense for Indic or other languages.
So you can imagine, in a market where voice based interfaces are quickly expanding across customer service, agriculture and public access systems, Bulbul V3 has become a functional requirement.
3. Context Over Cleverness
Sarvam AI performs more reliably than global AI models in scenarios that specifically involve local references, mixed language inputs and culturally specific phrasing. These situations are pretty common in everyday Indian usage, but they don’t always translate cleanly for global models.
There is no denying that ChatGPT and Gemini are amazing when it comes to general reasoning and broad problem solving, but when it comes to India specific context, Sarvam AI is really taking the lead.
Who Is Pratyush Kumar and Why His Leadership Matters
Sarvam AI was founded in 2023 by Pratyush Kumar and Vivek Raghavan. The aim was simple: build basic AI models that are designed for Indian languages and local use cases, rather than adopting global AI systems built for different markets.
Kumar is an IIT Bombay alumnus with an experience in AI research and applied systems. He has played a central role in helping shape much of the company’s early choices. But what stands out here is the direction in which Sarvam chose to move forward under his leadership.
The goal was never to run behind these fast moving global hype cycles, but to focus on areas that could be validated and tested like – benchmark performance, real deployments and India specific infrastructure constraints, that are pretty common in the Indian environment.
This approach sets Sarvam apart from a large number of AI startups
Why the World Is Suddenly Paying Attention
If you have been an AI enthusiast, you would know that the Global AI space has long been dominated by US based companies. So naturally, when an Indian model starts surpassing huge systems like ChatGPT and Google Gemini in niche but particular benchmarks, people start paying attention.
The emphasis for Sarvam AI has always been on how good it is at performing in the areas it was designed for, specifically for tasks that are related to local Indian languages, document complexity and regional context.
Now that focus, more than the broad claims of superiority, is what makes Sarvam AI stand out alongside ChatGPT, Google Gemini and other LLM’s.
Sarvam AI India and the Idea of Sovereign AI
But what is the idea behind Sovereign AI?
To put it simply, it refers to systems wherein the data stays local, the AI models are trained with keeping local contexts in mind, and the crucial infrastructure is not completely dependent on the foreign providers.
For India, this is more strategic than technical, because control over things like data, language models and system implementation, has a direct impact on the governance, scale and long term freedom in AI adoption.
Thus, by building an AI model that prioritises Indian language and local use cases, Sarvam AI is very much aligned with how sovereign AI is being applied in practice. And that is why policymakers, enterprises and global analysts are observing and following the progress being made in AI development, in countries outside China and the USA.
What This Means for Indian AI Startups
Sarvam AI’s rise offers a quiet but useful point of reference, for how smaller teams can also compete with Big techs. The company has totally focussed on getting things right in the few specific areas with clear local relevance, instead of focussing on matching the global platforms at every single possible capacity.
The Indian startups that are working in healthcare AI, language tech., governance systems and enterprise automations should focus more on depth and contextual accuracy rather than just scaling, as Sarvam is a prime example of how target focus can bring in meaningful results.
The Bigger Picture
It’s quite clear by now that Sarvam AI is not trying to position itself as a replacement for platforms like ChatGPT or Google Gemini. The focus for this one is completely local.
If the current direction continues with Sarvam AI, then it is very likely that it’s going to be viewed as an example of how localised AI development can compete with much larger platforms.
And that’s usually how the real disruption starts.
