In a Bengaluru event, Ola CEO Bhavish Aggarwal introduced India's very own AI model named Krutrim. This groundbreaking AI model demonstrated its prowess during the event, showcasing its ability to respond in diverse Indian languages and perform real-time coding.
Aggarwal emphasized Krutrim's unique ability to comprehend the Indian context, setting it apart from AI models predominantly trained on Western data.
Krutrim has made its debut in two classifications. The foundational LLM model, trained on an extensive dataset of 2 trillion tokens, boasts proficiency in understanding 22 languages and generating output in 10 Indian languages. Ola claims that Krutrim surpasses the performance of GPT-4 and Llama models, especially concerning Indic performance.
Ola asserts that Krutrim AI has undergone training on an extensive amount of Indic tokens, claiming a 20-fold increase compared to other models. A standout feature is its capability to comprehend voice commands and respond likewise.
The team behind Krutrim in Bengaluru hinted at a more robust model in development - Krutrim Pro. This advanced model is poised to be multi-modal, slated for release in the upcoming quarter.
Aggarwal emphasized that Krutrim has been meticulously built from the ground up, with the entire framework conceptualized by the team behind the AI. During a Q&A session, he highlighted the necessity for a solution designed specifically for India, aligning with Indian performance benchmarks, cultural relevance, and cost structures for India to lead as an AI-first economy.
Interested users can register their interest in the Krutrim AI model through the newly launched website. Registrations require a phone number, and the company pledges to notify users when the chatbot becomes available. Aggarwal confirmed the release of the base LLM model for everyone next month, with APIs expected to be accessible by February.
Key Takeaways:
Ola CEO Bhavish Aggarwal unveils Krutrim, India's first AI model built from the ground up for Indian context.
Krutrim boasts multi-lingual capabilities, understanding 22 languages and generating output in 10 Indian languages.
The base LLM model, trained on 2 trillion tokens, outperforms GPT-4 and Llama in Indic language tasks.
Krutrim Pro, a multi-modal model, is under development and expected for release next quarter.
The model accepts and responds to voice commands, further enhancing its user experience.
Access to the base LLM model begins next month, with APIs following in February.