A New Era: Indian Startups Embrace DeepSeek in AI

As the competition intensifies between OpenAI and DeepSeek, the landscape of artificial intelligence is experiencing a notable shift. Indian startups are increasingly weaving DeepSeek into their technological offerings. However, the scenario quickly evolved with the launch of OpenAI's o3-mini, a highly cost-effective reasoning model.

The introduction of China's DeepSeek-R1 has ignited a dynamic confrontation within the generative artificial intelligence (GenAI) sphere. As the industry braces itself for the outcome of this high-stakes rivalry, characterized as a David versus Goliath scenario, many are keenly watching the developments.

Competitive Edge: o3-mini vs. DeepSeek

Driven largely by the initial success of DeepSeek-R1, OpenAI made the strategic move to significantly reduce its model's pricing, enhancing its allure for potential adopters. Sam Altman, the visionary behind OpenAI, suggested in a Reddit dialogue that an open-source strategy could be on the horizon, though it's not yet their main focus. The pressing question remains: can o3-mini outperform DeepSeek R1?

Prasanna Krishnamoorthy, the managing partner at Upekkha, an AI fund and accelerator, observed the limited complimentary access to o3 is directly due to DeepSeek's influence. He pointed out that it doesn’t dramatically alter the enterprise landscape, which might prefer the open-source route with DeepSeek. He emphasized its advantages for individuals, especially if it's superior to o1-mini. On January 31, OpenAI’s o3-mini was publicly launched, surpassing o1-mini in disciplines like math and coding, all while maintaining a competitive pricing model.

Arko Chattopadhyay, co-founder & CEO of PipeShift, a model evaluation and training platform, commented on the critical juncture for enterprises: deciding between closed or open-source pathways.

Real-World Applications: Exploring DeepSeek's Utility

Chattopadhyay reflected on the practical applications, citing an e-commerce client using DeepSeek as its core model for reasoning and translations while maintaining Llama as a backup plan.

Meanwhile, an insurance-tech client previously piloting OpenAI’s o1 for document processing is now transitioning to DeepSeek to benefit from on-premise deployment capabilities, reducing reliance on OpenAI’s APIs.

Rashid Khan, co-founder and chief product officer at Yellow.ai, noted DeepSeek's potential in reducing GPU costs by 30-40%. He affirmed DeepSeek's significant margin of superiority over other open-source options, which benefits a wide array of sectors due to its open-source availability, from healthcare to government bodies.

OpenAI’s o3-mini is lauded for empowering developers with freedom in code creation and complex reasoning capabilities, Khan observed.

Abhishek Upperwal, founder of Soket Labs, illustrated how they leveraged DeepSeek for better performance in training and refining real-time speech APIs for Indic languages compared to Llama 70B, highlighting the advantageous balance of cost and capability DeepSeek offers.

Furthermore, Von Neumann Computers, a Bengaluru startup, is enabling its JOHNAIC server users to experiment with DeepSeek. Founder Sasank Chilamkurthy explained how distillation versions of DeepSeek’s comprehensive model outperform larger counterparts in computational tasks like math and coding.

Overcoming Barriers

Nevertheless, challenges remain for enterprises engaging with DeepSeek and similar Chinese models via APIs, primarily due to international security concerns. Italy and Australia, for instance, have expressed reservations. In Taiwan, government use of DeepSeek is prohibited due to security fears. Khan notes potential geopolitical sanctions could pose adoption hurdles, prompting hyperscalers like Microsoft Azure to quickly integrate DeepSeek on their platforms.