Bengaluru-based startup Sarvam AI is reshaping India’s AI landscape with its innovative “sovereign AI” models. Unlike the typical focus on the US and China, Sarvam is developing foundational AI tools such as Sarvam Vision and Bulbul, gaining significant recognition. Sarvam Vision outperforms established models like ChatGPT and Google Gemini in optical character recognition (OCR), achieving an impressive 84.3% accuracy on olmOCR-Bench and excelling on OmniDocBench v1.5, particularly with complex document layouts.
The startup’s focus on Indic-language models has shifted from skepticism to acclaim, with tech commentator Deedy Das acknowledging their superior text-to-speech (TTS) and OCR capabilities. Bulbul V3, the newly launched TTS model, supports over 35 voices across 11 Indian languages, aiming for broader inclusivity. Sarvam AI is proving that India can be a critical player in global AI development, filling significant gaps left by larger players in the industry.
Source link
