Sarvam AI’s Bengaluru-built Sarvam Vision posts an 84.3% OCR accuracy score, beating Gemini and ChatGPT while showcasing India’s sovereign, locally controlled AI capability for Indic languages.
India’s push for a sovereign, locally controlled AI stack has gained a decisive milestone with Sarvam AI launching Sarvam Vision, a homegrown model that has outperformed Google Gemini and OpenAI’s ChatGPT on specialised benchmarks, particularly Optical Character Recognition (OCR).
Sarvam Vision recorded 84.3% accuracy on the olmOCR-Bench, scoring higher than Gemini 3 Pro, DeepSeek OCR v2, and significantly above ChatGPT. The result places India among a small group of countries, alongside the US and China, with domestically developed foundation or LLM-class AI systems.
Positioned as a sovereign AI platform, Sarvam Vision is built in India using local talent and tailored for India-specific use cases, reducing dependence on foreign AI infrastructure. Its core strengths include Indic language AI, OCR, text-to-speech, and speech-to-text—areas where focused optimisation appears to outperform larger, general-purpose global models. OCR remains “one of the most test for an AI model,” making the benchmark a critical differentiator.
Pratyush Kumar, Co-founder, Sarvam Vision AI, shared the benchmark outcomes in a series of posts on ‘X’, highlighting the model’s comparative performance against ChatGPT and Gemini.
Industry observers have taken note. Deedy Das, tech commentator, said:
“I was wrong about Sarvam. When I wrote about them a year ago, I felt like the direction to train small Indic language models was wrong. But boy, have they turned it around. They have the best text-to-speech, speech-to text, and OCR models for Indic languages, and that’s actually really valuable. The pricing is very reasonable.”
The Bengaluru startup’s rise underscores India’s growing AI independence and the viability of regional-language, open and sovereign AI ecosystems.
