Chinese startup DeepSeek has unveiled a new experimental model, DeepSeek-V3.2-Exp, designed to boost efficiency and improve large-scale data handling at a fraction of the cost.
The launch, announced Monday on AI forum Hugging Face, builds on its current model DeepSeek-V3.1-Terminus and underscores the company’s mission to make AI systems faster and less resource-intensive.
DeepSeek first captured Silicon Valley’s attention last year with its surprise release of the R1 model, proving it was possible to train large language models rapidly on less powerful chips with fewer resources, according to CNBC.
“It’s significant because it should make the model faster and more cost-effective to use without a noticeable drop in performance,” said Nick Patience, vice president and practice lead for AI at The Futurum Group.
“This makes powerful AI more accessible to developers, researchers, and smaller companies, potentially leading to a wave of new and innovative applications.”
An AI model makes decisions based on both its training data and new inputs, such as a prompt.
Consider an airline trying to determine the best route from point A to point B. While there may be many possible paths, not all are practical. By filtering out the less viable options, the airline can save significant time, fuel, and money.
That’s essentially what sparse attention does. Instead of processing all the available data like traditional models, it selectively focuses only on the information most relevant to the task at hand, making the system far more efficient.
“So basically, you cut out things that you think are not important,” said Ekaterina Almasque, the cofounder and managing partner of new venture capital fund BlankPage Capital.
DeepSeek insists the experimental model matches the performance of V3.1-Terminus.
Even as some warn of a looming AI bubble, the sector remains central to the geopolitical contest between the U.S. and China for technological supremacy.
While the latest version aims to push those boundaries further, experts caution that questions remain about the effectiveness and safety of its architecture.
