India's Sarvam-2B beats GPT-4o
Sarvam AI has just made history by launching India's first open-source foundational model in 10 Indic languages, Sarvam 2B. Sarvam 2B powerful Indic language model is trained on a massive 4 trillion tokens, with 2 trillion tokens dedicated to Indic languages alone. Sarvam AI’s results are astonishing and a bit scary for GPT-4o. Sarvam 2B is outperforming GPT-4o in accuracy and speed, challenging global giants like Google's Gemma 2-9B and Meta's Llama3.1-8B models. But apart from beating GPT-4o, Sarvam AI has also unveiled Shuka, an innovative audio language model tailored for Indic audio, opening new frontiers for voice-based applications in India. Sarvam AI’s enterprise solutions, such as Sarvam Voice Agents, are already making an impact, processing over 27,000 transactions for Sri Mandir, a rising startup.