Until now, the world of artificial intelligence (AI) was mainly dominated by the US and China.
India was often seen as a talent pool rather than a center for core AI innovation. However, Bengaluru-based startup Sarvam AI is changing this image.
The company is now challenging global tech giants with its sovereign AI model, which has been developed entirely in India.
Sarvam Vision: A Powerful AI Model
Sarvam AI’s two tools, Sarvam Vision and Bulbul, are currently gaining attention.
Sarvam Vision is an OCR (Optical Character Recognition) based AI model that has outperformed major AI models like ChatGPT, Google Gemini, and Anthropic Cloud on some benchmarks.
Its accuracy is so high that users and AI experts are openly praising it.
Sarvam AI co-founder Pratyush Kumar shared these achievements in a post on X. According to the company, Sarvam Vision scored 84.3% accuracy on olmOCR-Bench, which is higher than models like Gemini 3 Pro and DeepSeek OCR v2. In comparison, ChatGPT scored much lower.
Sarvam Vision also achieved an impressive 93.28% score on OmniDocBench v1.5. It performed especially well in difficult areas such as complex layouts, technical tables, and mathematical formulas, where traditional OCR systems usually struggle.
From Doubt to Recognition
Initially, Sarvam AI faced criticism because it focused mainly on Indic language models. But over time, this criticism has turned into appreciation.
Tech commentator Deedy Das admitted that he had underestimated Sarvam. He said Sarvam’s OCR and speech models are very strong for Indian languages
and fill a gap that global AI labs had ignored. Many users have also shared their surprise and excitement after using Sarvam’s tools.
Bulbul V3: AI Voice for Indian Languages
Along with OCR technology, Sarvam AI has launched its new text-to-speech model, Bulbul V3. This AI voice tool is designed to create natural and expressive voices in Indian languages.
Its concept is similar to global AI voice platforms like ElevenLabs, but it is built specifically for India’s needs.
Currently, Bulbul V3 offers more than 35 voices in 11 Indian languages, and the company plans to expand it to 22 languages soon.
