Universal AI: Till now, only America and China were considered to have dominance in the world of Artificial Intelligence. India is often considered only a market for talent and not a center for core AI development. But Bengaluru-based startup Sarvam AI is rapidly changing this thinking. The company is now giving a direct challenge to the global tech giants on the basis of its sovereign AI model developed from the beginning in India.
Sarvam Vision left big AI models behind
Two tools of Sarvam AI are in discussion these days, Sarvam Vision and Bulbul. Sarvam Vision is an OCR (Optical Character Recognition) based AI model that has outperformed big names like ChatGPT, Google Gemini and Anthropic Claude on some select benchmarks. Its accuracy is said to be so good that users and AI experts are openly praising it.
Benchmark score increased the world’s interest
Sarvam AI co-founder Pratyush Kumar informed about these achievements through posts on X. According to the company, Sarvam Vision has achieved an accuracy of 84.3 percent on olmOCR-Bench, which is higher than models like Gemini 3 Pro and DeepSeek OCR v2. Whereas ChatGPT’s score was much lower than this.
Not only this, Sarvam Vision also scored an excellent score of 93.28 percent on OmniDocBench v1.5. The special thing is that its performance was very strong even in difficult areas like complex layout, technical tables and mathematical formulas where traditional OCR systems often fail.
Journey from criticism to praise
Sarvam AI was initially viewed with suspicion for focusing only on Indic language models. But now that doubt is turning into appreciation. Tech commentator Deedy Das also admitted that he had underestimated Sarvam. According to him, Sarvam’s OCR and speech models are extremely robust for Indian languages and fill a gap that was ignored by large global AI labs. The reactions of the users are also similar. Many people have expressed both surprise and enthusiasm after using Sarvam’s tools.
Bulbul V3
Along with OCR, Sarvam AI has also launched its new text-to-speech model Bulbul V3. This AI voice tool focuses on creating natural and expressive voices in Indian languages. Its concept is similar to international AI voice platforms like ElevenLabs but it has been specially designed according to India-centric needs. Currently, Bulbul V3 has more than 35 voices available in 11 Indian languages and the company plans to expand it to 22 languages soon.
Also read:
Ads will not be visible on YouTube! This method is amazing, 99% people do not know this jugaad
