A team of researchers in the US claim to have reproduced Deepseek’s viral AI chatbot for just $30.
The Chinese AI startup caused a trillion-dollar dent to the US tech industry last week after its latest R1 model outperformed OpenAI’s ChatGPT – despite being built at a fraction of the cost.
DeepSeek said it created its AI model for less than $6 million, using hardware that is inferior to US rivals like OpenAI and Meta.
Now, AI researchers at the University of California, Berkeley, claim to have replicated the R1 model for under $30.
Dubbed TinyZero, the new AI model was trained using the number puzzle game found in the game show Countdown, which requires players to reach a predetermined answer from a random set of numbers.
“We reproduced DeepSeek R1-Zero in the CountDown game, and it just works,” Berkeley PhD student Jiayi Pan, who led the research, wrote on X.
“And it costs <$30 to train the model. We hope this project helps to demystify the emerging RL scaling research and make it more accessible!”
The research, which is yet to be peer-reviewed, is available on the developer platform GitHub.
Jiayi Pan noted that the model would need far more computational power to be validated in the general reasoning domain, which would cost much more than $30.
Despite its limitations, the ability to recreate an advanced AI model for such a small cost has once again raised questions about the current funding models of US tech giants, who have spent hundreds of billions of dollars developing their most advanced models.
Industry experts say that DeepSeek’s success may have triggered a shift in the way such models are built and used in the future.
“The shift has begun,” James Fischer, chief strategy officer at data analytics firm Qilk, said following the release of DeepSeek’s R1 model.
“The costs of running advanced AI models are dropping dramatically, levelling the competitive playing field. This, in turn, pushes AI into its next phase, away from the infrastructure-heavy focus of training and into Applied AI – the era of putting AI to work in practical, scalable ways.”
