OLA’s founder, Bhavish Aggarwal, has invested $ 230 million in AI startup established to establish the country in the field of US and Chinese companies.
Aggarwal provides money to invest in Krutrim, which is building LLM for Indian languages, mainly through his family office. At the X Tuesday post, Aggarwal said that Krutrim is aiming to invest $ 1.15 billion by next year. He said that he was trying to procure the remaining capital from an outside investor.
Financial announcements are the same as announcing plans that Unicorn’s startup Krutrim has an open source of AI models and to build a partnership with NVIDIA as India’s largest supercomputer.
The lab has released the Krutrim-2, a 12 billion parameter language model that showed a powerful performance in the processing of Indian language. In the emotional analysis test shared by Krutrim on Tuesday, the competitive model gained 0.95 compared to 0.70 and achieved a 80 % success rate in code generation tasks.
Labs have several specialized models, including images processing systems, audio translation, and text search systems, all optimized for Indian languages.
“We are not close to a global benchmark yet, but we have made a good progress in a year,” says Aggarwal, who is supporting other ventures by SoftBank in X. Create a world -class Indian AI ecosystem. “
This is because India is trying to establish himself in an artificial intelligence landscape dominated by the United States and Chinese companies. The recent release of DeepSeek’s R1 “Progress” model, built based on a modest budget, has sent a shock wave through the high -tech industry.
India praised the progress of Deepseek last week, and said that the country would host a large -scale language model of Chinese AI labs on a domestic server. KRUTRIM’s cloud arm started providing DeepSeek on an Indian server last week.
Krutrim has also developed Bharatbench, a unique evaluation framework that evaluates the proficiency of the AI model in Indian, and is dealing with existing benchmark gaps mainly focusing on English and Chinese.
Lab’s technical approach includes a 128,000 -toe context window, so that the system can process long textbooks and more complex conversations. The performance metrics issued by the startup showed Krutrim-2, which achieves a high score in grammar correction (0.98) and multi-turn conversation (0.91).
This investment continues to the release of Krutrim-1 in January. This is the 7 billion parameter system that functions as India’s first large language model. NVIDIA’s supercomputer deployment is scheduled to be released in March, and expansion is planned throughout the year.
Source link
