New Delhi: A little-known Indian artificial intelligence start-up has vaulted into the global spotlight after its models outperformed ChatGPT and Google Gemini in select India-focused benchmarks, underlining the growing strength of homegrown innovation in the AI race.
Sarvam AI, founded in 2023 by Indian researchers Pratyush Kumar and Vivek Raghavan, has developed AI systems tailored for India’s linguistic diversity and real-world document complexity. Its document understanding and optical character recognition capabilities have delivered higher accuracy in reading multi-script Indian documents, forms and tables, an area where large global models often falter.
Equally striking is Sarvam’s text-to-speech engine, Bulbul V3, which generates natural-sounding voices across several Indian languages and handles code-mixed speech such as Hinglish with ease. The model has drawn praise for its stability, clarity and cost efficiency, making it suitable for large-scale public and enterprise use.
The breakthrough has also caught the attention of policymakers. Amitabh Kant, former CEO of NITI Aayog, described Bulbul V3 as “pathbreaking AI” and a strong example of sovereign, indigenous technology built for India’s cultural and linguistic realities. His endorsement has added momentum to the growing view that India can build globally competitive AI without relying solely on foreign platforms.
Sarvam AI’s approach is sharply focused, building specialised models optimised for Indian use cases rather than chasing sheer scale. While experts caution that its success is task-specific and does not yet rival global giants across all AI capabilities, many see it as proof that context-aware, locally trained systems can outperform larger models where it matters most.
As debates around data sovereignty and digital independence intensify, Sarvam AI’s rise signals a shift in India’s technology story - from being a consumer of global AI tools to emerging as a serious creator of them.
BI Bureau
