India’s 20 LLMs vs China’s DeepSeek

India is finally making big moves in the AI space. With ?2,000 crore allocated in the Union Budget 2025 for the IndiaAI mission, the government invited startups, researchers, and entrepreneurs to collaborate on AI models trained on Indian datasets.?

And the proposals have rolled in.

By Mohit Pandey

According to reports, the government has received 67 proposals, 20 of which are focused on building LLMs. Big names like Sarvam AI, CoRover.ai, and Ola Krutrim are in the mix, aiming to create India’s homegrown AI models.

A senior government official has confirmed that the electronics and IT ministry is setting up a high-level technical committee to evaluate the proposals. This expert panel will decide which projects get the green light within a month.

One of the most interesting submissions was a sector-specific small language model created by a group of doctors to assist medical professionals working on breast cancer.

With the proposals in and the government keen on moving quickly, India’s foundational AI models could be kicking off soon.

Meanwhile, DeepSeek is going all-in on open-source AI.

In China, DeepSeek AI is doubling down on open-source AI models. The team announced on Friday that five open-source repositories will be launched next week as part of their “open-source week”.

“We’re a tiny team exploring AGI,” the company wrote on X, promising to share its progress with full transparency.

DeepSeek already has 14 open-source models and repositories on Hugging Face, and its recent DeepSeek-R1 and DeepSeek-V3 are proving to be state-of-the-art models trained on far fewer resources than their competitors.

With India gearing up to build its foundational AI models and DeepSeek leading the open-source charge, the global AI landscape is shifting fast. However, there’s a faction that firmly believes we should not be chasing foundational models.?

At Razorpay FTX’25, CRED co-founder and CEO Kunal Shah said India should treat AI models like WhatsApp and LED bulbs. “We didn’t question why India didn’t invent WhatsApp or LED bulbs. We just adopted them and became the best at using them. That should be our approach with AI,” Shah argued.

“We built UPI; we are the beneficiaries of that. The world should be copying that.” Shah said the question we should be asking is how to leverage what we already have. He added that most jobs in India are hyper-inefficient, and as a large country, we should be finding ways to solve that using AI.

Click here to check out the full story.

AI Bytes

• TCS has announced a collaboration with Salesforce to enhance the use of AI in the manufacturing and semiconductor sectors.
• Arize AI, a company focusing on AI observability and LLM evaluation, has announced a remarkable $70 million Series C funding round.
• CTGT, a YC-backed AI startup, has secured $7.2 million to drive its mission of scaling AI beyond traditional deep learning.
• Microsoft has unveiled a new deep-learning model called Biomolecular Emulator-1 (BioEmu-1), which can generate thousands of protein structures hourly.
• California-based Together AI has announced a successful $305 million Series B funding round, led by General Catalyst and co-led by Prosperity7.