A team of researchers from UC Berkeley claims to have successfully replicated a small-scale version of DeepSeek R1-Zero for just $30, raising questions about the true cost of AI development.

The revelation fuels ongoing debates over the pricing and valuation of AI technology by major companies.

OpenAI, for example, currently charges $15 per million tokens via its API, while DeepSeek offers a significantly lower rate of $0.55 per million tokens. The Berkeley team’s findings suggest that highly capable AI models could be built at a fraction of the cost currently invested by industry leaders.

Jiayi Pan, a PhD candidate leading the project, introduced TinyZero — an open-source AI model available for public experimentation and training. While still in its early stages, TinyZero has demonstrated its ability to play Countdown and perform multiplication tasks, showcasing adaptability in strategic problem-solving.

The research challenges a fundamental assumption in AI: Is artificial intelligence truly as resource-intensive as tech giants claim, or is the industry inflating costs to maintain control over its development?

DeepSeek, a rising Chinese AI company, has been making waves in the industry, triggering massive shifts in the U.S. tech market. The Berkeley team’s discovery could further challenge the status quo, potentially reshaping the economics of AI development.