Ola CEO Bhavish Aggarwal today unveiled “Krutrim,” India’s own Artificial Intelligence (AI) model. In a launch event held today in collaboration with Krutrim Si Designs, an AI company which he has co-founded, Aggarwal showcased the capabilities of Krutrim, emphasising its unique features tailored for the Indian context.

Krutrim comes in two classifications: the base Large Language Model (LLM) and the more powerful multi-modal Krutrim Pro, set to be available next quarter. The base model has been trained on an impressive 2 trillion tokens, understanding 22 languages, and generating output in 10 Indian languages. Krutrim Pro, with advanced capabilities, is poised to further elevate AI experiences.

Aggarwal asserted that Krutrim outperforms GPT-4 and Llama models, specifically excelling in Indic performance. The model can engage in real-time coding and voice interactions, addressing the need for AI models grounded in the Indian cultural context. Aggarwal however, did not shed any light on how the model has actually been built, considering even the biggest of corporations globally have taken months to come with a barely functioning LLM. Add to that the GPU challenges that the sector is facing, and it becomes clearer on how building LLMs from scratch is no mean task. From the looks of it, Krutrim looks strikingly similar to ChatGPT on UI/UX etc., and could actually well be built over an existing API. No clarity was provided by the company on that either.

The Krutrim website is now live for registrations, with the base LLM becoming available next month. APIs are anticipated in February 2024, while Krutrim Pro is slated for release in the following quarter.

Krutrim Si Designs, founded in April 2023 by Bhavish Aggarwal and Krishnamurthy Venugopala Tenneti, positions Krutrim as India’s first full-stack AI solution. The company emphasises the importance of an AI model rooted in local Indian knowledge, languages, and data. It however does not make it clear as to what is the base for building such a complicated, tough indic-language AI LLM. No research papers or datasets have been released so far.

The unveiling of Krutrim follows recent developments in the AI space, including OpenHathi and Google’s Gemini. Aggarwal envisions Krutrim as an AI company tailored for Indian customers, leveraging the nation’s abundant data resources.

The launch event featured a live demonstration showcasing Krutrim’s ability to comprehend and generate content in multiple Indian languages. A comparison with other AI models highlighted Krutrim’s superior performance in Indic languages. Insight into the team behind Krutrim Si Designs and the foundational model training landscape was also provided.

Interested users can sign up on the Krutrim website for early access, with an open release scheduled for January 2024. Krutrim APIs will be accessible to developers in February 2024, and Ola group companies are set to integrate Krutrim across various functions from March 2024.

Krutrim’s launch makes India’s nascent but rapidly developing generative AI space exciting. It comes at a time when OpenAI itself is targeting an India entry, tying with former Twitter India boss for a planned roll-out. India also recently concluded a massive global AI conference, which aims at responsible use of AI and includes signatories such as the US and almost the entire of Europe.