Sarvam launches indigenous AI language models, targets global standards

The Indian startup unveiled two large language models built for 22 Indian languages, claiming competitive performance with global AI systems while offering lower inference costs and enhanced voice-first capabilities for diverse use cases.

Indian artificial intelligence startup Sarvam AI has introduced two homegrown large language models (LLMs) designed specifically for Indian languages, marking a significant step in the country’s push toward AI self-reliance. The newly launched models feature 30 billion and 105 billion parameters and are built to support multilingual and voice-first applications.

According to the company, the 30-billion-parameter model is optimised for real-time conversational use with a 32,000-token context window, helping reduce inference costs. The larger 105-billion-parameter model, equipped with a 128,000-token context window, is aimed at more advanced reasoning and complex problem-solving tasks.

Built for complex reasoning and Indian contexts

Sarvam’s co-founder Pratyush Kumar said the 105-billion-parameter model performs competitively with leading global systems in areas such as mathematical reasoning, coding accuracy and analytical problem-solving. He noted that despite being significantly smaller than some international models, it delivers comparable intelligence while remaining cost-efficient.

At the launch event, the company demonstrated its AI chatbot “Vikram,” which conversed in multiple Indian languages including Hindi and Punjabi. The chatbot is named in tribute to Vikram Sarabhai, the renowned Indian physicist and space pioneer.

The models support all 22 officially recognised Indian languages and are tailored for voice-led interactions — a key focus area as AI adoption expands beyond text-based interfaces. The company believes voice-enabled AI will accelerate digital inclusion across India.

Part of India’s broader AI push

Sarvam AI has been selected under the India AI Mission to help develop a sovereign LLM ecosystem, including an open-source 120-billion-parameter model aimed at enhancing governance and public services. Other Indian firms, including Soket and Gnani, have also announced large-scale foundation models targeting sectors such as defence, healthcare and education.

The launch comes amid growing competition in India’s AI landscape. Global players like OpenAI and Anthropic have expanded support for Indian languages, underscoring the country’s importance as a major AI market.

Sarvam has reportedly raised around $50 million from investors including Khosla Ventures, Lightspeed Venture Partners and Peak XV, positioning itself as a key contender in India’s evolving AI ecosystem.