Google's AI boss casts doubts on DeepSeek's efficiency claims, calls them exaggerated and misleading

Demis Hassabis, CEO of Google DeepMind, has expressed doubts over recent claims made by Chinese AI startup DeepSeek regarding its model’s cost efficiency. Speaking at the Artificial Intelligence Action Summit in Paris, Hassabis acknowledged DeepSeek’s accomplishments but described its reported cost-efficiency figures as “exaggerated and a little bit misleading.”

DeepSeek recently gained global attention with its R1 model, which it claims can rival OpenAI’s GPT-4-level o1 model despite being trained for just $5.6 million—a fraction of the $100 million-plus that OpenAI reportedly spent on GPT-4. However, Hassabis argued that the $5.6 million likely reflects only the final training run, omitting substantial costs tied to development, including data collection, infrastructure, and multiple iterations of training.

Hassabis also hinted that DeepSeek may have relied heavily on knowledge gained from existing Western AI models to develop its own. OpenAI has previously raised similar concerns, alleging that companies from China often distil or replicate the work of leading US-based AI firms. Following DeepSeek’s launch, OpenAI told Bloomberg that companies based in China and other countries frequently attempt to extract and refine Us AI models for their own use.

More from Tech

Govt. of India likely to issue formal advisory against using DeepSeek over data security concerns Will 2025 be another year of layoffs?

Google doesn’t view DeepSeek as a game-changer

Despite the buzz surrounding DeepSeek, Hassabis downplayed the notion that it represents a major leap in AI efficiency. He claimed that Google’s Gemini models offer better cost-to-performance efficiency, even if they haven’t been promoted with the same level of marketing fanfare. “It’s impressive, but it isn’t some new outlier on the efficiency curve,” Hassabis remarked.

DeepSeek’s claims have sparked lively debate across the tech world, with experts divided on whether the startup has genuinely redefined the cost of AI development or simply found a clever way to market itself.

The real test: Innovation and long-term impact

While DeepSeek’s reported efficiency figures are impressive on paper, industry leaders like Hassabis suggest that true success in AI comes down to long-term innovation and reliability rather than bold claims. As the competition for more powerful and cost-effective AI models heats up, both established players like Google and emerging startups like DeepSeek will face growing scrutiny over their performance and transparency.

Ultimately, the ongoing race for AI dominance may hinge on which companies can consistently deliver breakthrough technologies without cutting corners—something both investors and governments are watching closely.