Sarvam AI Advances Indigenous Language Models

1.      Sarvam AI Advances Indigenous Language Models

Sarvam AI, a Bengaluru-based startup founded by Vivek Raghavan and Pratyush Kumar, has emerged as a major success story under the IndiaAI Mission, earning praise from Union Ministers Dharmendra Pradhan and Ashwini Vaishnaw. The company focuses on developing AI models deeply aligned with Indian languages, contexts, and diversity. Its flagship 24-billion-parameter Sarvam-M large language model demonstrates strong reasoning in Indian languages, mathematics, and coding, while Sarvam-1 (2B parameters) is optimized for low-power devices. Other innovations include Saaras V3 (speech-to-text supporting all 22 scheduled languages), Bulbul V3 (emotion-aware text-to-speech in 35+ voices), Sarvam Vision (OCR for complex multilingual documents), Sarvam Dub (AI-based dubbing retaining speaker identity), and Sarvam Arya (multi-agent AI orchestration platform). Notably, Sarvam Vision achieved 84.3% accuracy on olmOCR-Bench and 93.28% on OmniDocBench v1.5, outperforming global models like Gemini and DeepSeek despite using only 3B parameters, exemplifying frugal innovation. A 70B-parameter model is under development. Supported with 4,000 GPUs by the government, Sarvam AI strengthens India’s data sovereignty, promotes inclusive digital access in regional languages, and positions India as a competitive force in the rapidly expanding global AI ecosystem.