India's Sarvam AI Surpasses Google and ChatGPT in Innovation

Sarvam AI Indian AI models Artificial Intelligence India LLM India Sarvam Samvaad Bulbul V3 Sarvam Vision
Deepak-Gupta
Deepak-Gupta

CEO/Cofounder

 
February 9, 2026
3 min read
India's Sarvam AI Surpasses Google and ChatGPT in Innovation

TL;DR

  • Sarvam AI is developing AI models specifically for Indian languages and use cases, focusing on compact and efficient designs for local deployment. Their recent launches, Bulbul V3 and Vision, demonstrate competitive performance against global AI systems on India-focused benchmarks, showcasing promising advancements for the region's technological landscape.

Sarvam AI: India's Answer to Global AI Models

Sarvam AI, based in Bengaluru, is developing language and voice models designed for Indian languages and use cases. Founded in 2023 by Dr. Vivek Raghavan and Dr. Pratyush Kumar, Sarvam AI focuses on creating compact, efficient models suitable for phones, call systems, and local languages. The company's goal is to build AI that works effectively within India's linguistic diversity and bandwidth constraints.

Sarvam AI's Approach to AI Development

Sarvam AI emphasizes careful data curation and task tuning to outperform larger, generic models on India-specific problems. Their product line includes small-to-medium sized language models, speech tools, and APIs for speech-to-text and text-to-speech. The firm argues that this focused approach can yield better results for specific tasks within the Indian context.

Outperforming Global AI Models

Sarvam AI's recent launches, including Bulbul V3 (a text-to-speech system) and Vision (an OCR/vision model), have demonstrated competitive performance against global systems on targeted, India-focused benchmarks.

Sarvam AI

Image courtesy of India Today

Bulbul V3

Bulbul V3 recorded lower error rates on telephony-grade audio and better handling of numerals, named entities, and code-mixed text compared to several global TTS systems. These results are based on blind listening studies and automated error tests.

Vision

Early tests on document reading in Indian languages showed Sarvam AI’s Vision tool outperforming generalist models on some India-language OCR tasks. According to Pratyush Kumar, co-founder of Sarvam AI, Sarvam Vision achieved an accuracy score of 84.3 percent on the olmOCR-Bench, surpassing Gemini 3 Pro and DeepSeek OCR v2. The tool also scored 93.28 percent on OmniDocBench v1.5, demonstrating strong results on complex layouts, technical tables, and mathematical formulas.

Sarvam AI OCR

Image courtesy of India Today

Independent Verification and Significance

While Sarvam AI has presented data from blind listening studies and automated comparisons, it is important to note that vendor-led evaluations benefit from external replication to confirm rankings. Media reports suggest independent listener votes and large sample sizes for some tests. These results are notable as early evidence of Sarvam AI's capabilities.

Practical Applications and Future Prospects

Sarvam AI's tools offer potential benefits for Indian firms and public services, including cheaper, local-language voice agents and improved OCR for native scripts. The company collaborates with cloud partners and participates in government discussions on sovereign AI, potentially leading to increased adoption in government and telecom applications.

Sarvam AI Bulbul V3

Image courtesy of India Today

Bulbul V3 Details

Bulbul V3, Sarvam AI's text-to-speech AI model, is designed to generate natural and expressive voices for Indian languages. It supports 35+ voices across 11 Indian languages, with plans to expand to 22 languages.

Sarvam Samvaad: AI Agents for India

Sarvam AI offers Sarvam Samvaad, a platform to build, customize, and launch AI Agents tailored for India. These agents support 11 Indian languages and can be deployed across various channels, including phone calls, WhatsApp, web, and apps.

Sarvam AI Samvaad

Image courtesy of Sarvam AI

Sarvam Samvaad provides insights from every interaction, allowing users to track agent performance and analyze conversations.

Building with Sarvam AI

Sarvam AI enables users to create custom AI products and applications using Sarvam Models. The Government of India has selected Sarvam AI to develop India's sovereign foundational model.

Sarvam AI Build

Image courtesy of Sarvam AI

Latest Research from Sarvam AI

Sarvam AI is actively involved in AI research, with projects like:

  • Sarvam-M: A hybrid Indic model fine-tuned for Indian languages and reasoning tasks.
  • Sarvam Translate: An open-weights model for text translation across 22 Indian languages.
  • Sarvam-1: India’s first LLM for 10 Indian languages, trained on 2B parameters.

Explore how company name can help your organization leverage AI for enhanced efficiency and innovation. Contact us today to learn more about our comprehensive suite of services.

Deepak-Gupta
Deepak-Gupta

CEO/Cofounder

 

Deepak Gupta is a technology leader and product builder focused on creating AI-powered tools that make content creation faster, simpler, and more human. At Kveeky, his work centers on designing intelligent voice and audio systems that help creators turn ideas into natural-sounding voiceovers without technical complexity. With a strong background in building scalable platforms and developer-friendly products, Deepak focuses on combining AI, usability, and performance to ensure creators can produce high-quality audio content efficiently. His approach emphasizes clarity, reliability, and real-world usefulness—helping Kveeky deliver voice experiences that feel natural, expressive, and easy to use across modern content platforms.

Related News

Google Launches Gemini 3.1 Flash with Advanced TTS Capabilities for Enterprise Voice Infrastructure

Google Launches Gemini 3.1 Flash with Advanced TTS Capabilities for Enterprise Voice Infrastructure

Google Launches Gemini 3.1 Flash with Advanced TTS Capabilities for Enterprise Voice Infrastructure

By Ankit Agarwal April 27, 2026 4 min read
common.read_full_article
2026 Enterprise AI Update: GPT-4.1 and Llama Benchmarks Signal Shift in Multimodal Voice Infrastructure

2026 Enterprise AI Update: GPT-4.1 and Llama Benchmarks Signal Shift in Multimodal Voice Infrastructure

2026 Enterprise AI Update: GPT-4.1 and Llama Benchmarks Signal Shift in Multimodal Voice Infrastructure

By Ankit Agarwal April 24, 2026 4 min read
common.read_full_article
Amazon Commits $200 Billion to Scaling Multimodal AI Infrastructure for Enterprise Voice and Synthetic Media

Amazon Commits $200 Billion to Scaling Multimodal AI Infrastructure for Enterprise Voice and Synthetic Media

Amazon Commits $200 Billion to Scaling Multimodal AI Infrastructure for Enterprise Voice and Synthetic Media

By Ankit Agarwal April 20, 2026 4 min read
common.read_full_article
New Appinventiv Report Details Critical Biometric Authentication Risks in Enterprise AI Voice Cloning Systems

New Appinventiv Report Details Critical Biometric Authentication Risks in Enterprise AI Voice Cloning Systems

New Appinventiv Report Details Critical Biometric Authentication Risks in Enterprise AI Voice Cloning Systems

By Ankit Agarwal April 17, 2026 4 min read
common.read_full_article