There had been concerns that India was rapidly falling behind in the AI race, and there’s now acknowledgement from India’s most valuable AI company that is indeed the case.
Ola Krutrim has unveiled its latest set of AI models. These models have been made open-source, and Ola Krutrim has also published some technical details around the models. But these models don’t come anywhere close to the models released by top US AI labs, or even Chinese AI company DeepSeek.
“We’re nowhere close to global benchmarks yet but have made good progress in 1 year,” Ola Krutrim CEO Bhavish Aggarwal posted on X. “And by open sourcing our models, we hope the entire Indian AI community collaborates to create a world class Indian AI ecosystem. We’re still learning to walk before we can run, hopefully within this year!” he added.
Ola released a slew of new models, including the Krutrim 2 LLM, vision model Chitrarth 1, speech language model Dhwani 1 and a translation model Krutrim translate 1. But these models fell well short of the current benchmarks results from top models. For instance, Krutrim 2 performed worse on most benchmarks compared to the MN-12b-Instruct model, which was trained on top of France-based Mistral’s NeMo-12b model, which is open-source. MN-12b-Instruct was released by a seemingly small team in July 2024, but Krutrim 2 failed to match it on most benchmarks. Krutrim 2 also fell well short of Google’s Flash and GPT 4o models, which are already a generation behind OpenAI’s o1 and o3-mini series of models.
These don’t seem to be encouraging signs. Ola Krutrim’s models seem to be unable to beat the performance of models cheaply trained on open-source models released from French labs eight months ago. Eight months is a lifetime in AI development, and the gap between Krutrim and top US and Chinese labs is even greater — at this point, it appears that US and even Chinese companies might have built an insurmountable lead over India’s top AI startup.
Also, it appears that Indian companies like Ola Krutrim and Sarvam AI focused on the narrow task of building AI models for Indic use-cases and Indian languages. This might end up being inconsequential in the long run — with US and Chinese companies rapidly moving towards developing AGI, it’s likely that their models will outperform any language or region specific efforts that Indian companies might have. As such, Indian companies — like the Chinese — might be better served in looking to build general purpose models that the world can use, instead of focusing their efforts on Indian use-cases.
The only silver lining here is that Ola Krutrim seems to realize how far behind it is. The company has already raised $50 million from investors at a valuation of $1 billion, but has said it’ll invest another Rs. 2,000 crore into research this year, and a further Rs. 10,000 crore by next year. It remains to be seen if these efforts will be enough to help Indian companies catch up in the AI race, but based on Krutrim’s latest announcements, they certainly have a massive amount of catching up to do.