Sarvam AI Launches 22 Specialized Models Supporting India's Diverse Local Languages for Sovereign AI Revolution

Sarvam AI, a leading Indian AI startup, has officially launched 22 advanced language models optimized for India's 22 scheduled local languages.

These models, including enhancements to Sarvam Audio, Saaras V3, and Bulbul V3, provide superior performance in speech recognition, text-to-speech, and vision tasks tailored to Indic scripts and accents.

Bridging the AI Accessibility Gap in India

The launch marks a significant step toward making AI inclusive for over 1.4 billion Indians by supporting voice commands and real-time processing in native tongues.

Founded in Bengaluru, Sarvam AI has a rich history of innovation, starting with India's first homegrown multilingual LLM, Sarvam-1, trained on sovereign data.

The company has raised substantial funding and collaborated with institutions like IIT Madras and AI4Bharat to build massive datasets such as IndicVoices with 12,000 hours of speech data.

Outperforming global leaders like ChatGPT and Google Gemini in Indic OCR, document understanding, and noisy audio environments, these models set new benchmarks.

The impact extends to enterprises, media, government, and feature phones, enabling low-cost telephony at Rs 30 per hour and multilingual dubbing.

Future Roadmap and Sovereign AI Vision

Looking ahead, Sarvam plans to release larger 30B and 105B parameter models to compete directly with international AI giants.

Expansion to more voices, accents, and cross-lingual capabilities will further democratize AI across regions.

This initiative not only preserves cultural nuances but also positions India as a frontrunner in sovereign AI development.