Sarvam AI has launched two indigenous large language models with 30B and 105B parameters at the Indi...
The AMW Read
The launch of large-scale, language-specialized models by Sarvam AI validates the Frame 2 (CN/OSS challenger) pattern of regionalized, cost-effective competition against global frontier labs, specifically within the context of India's sovereign AI mission.
Sarvam AI has launched two indigenous large language models with 30B and 105B parameters at the India AI Impact Summit 2026, directly challenging Google, OpenAI and Anthropic in one of the world's fastest-growing AI markets. Both models support all 22 official Indian languages, use mixture-of-experts architecture, and are optimized for voice-first interaction, with the 105B model featuring a 128,000-token context window. The company claims its 105B model outperforms DeepSeek R1 and Google's Gemini Flash on key benchmarks while being cheaper to run, demonstrating that regionally-specialized open-source models can compete effectively with global giants. Backed by $50M+ from Lightspeed and Khosla Ventures and supported by India's sovereign AI mission, this marks a significant milestone in India's push for AI independence and signals a broader trend toward localized, language-specific AI infrastructure across emerging markets. 🇮🇳
