DeepSeek Unveils Open-Source Tool to Help Cut AI Operating Costs

DeepSeek is opening up the software it uses to run its artificial intelligence models at scale, potentially making its technology cheaper and easier for businesses to run.

The service, called DeepInfer, is software DeepSeek uses to make its models run faster and use computing resources more efficiently.

“We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable,” the Hangzhou-headquartered Chinese AI company said on Tuesday on X.

DeepInfer supports DeepSeek’s V4.1-Flash model for now, but support for future systems will follow, the company said.

The release casts light on how DeepSeek keeps down the cost of actually running advanced AI, known as inference, offering a glimpse into production techniques that AI companies typically keep closely guarded.

It also offers a way for cloud and hosting companies running DeepSeek’s models to deploy those cost-cutting methods themselves.

The same model can cost substantially more to run, depending on the software used to serve it. More efficient software can lower the computing cost of each request and potentially reduce energy use.

DeepInfer brings together different techniques such as distributed inference — spreading AI workloads across different hardware — that allow models to produce more output in less time and squeeze more performance from chips. Many of the underlying techniques are widely used, but DeepSeek has tailored them to work efficiently with its models.

Lower inferencing costs — the costs of running AI models — can improve margins for AI hosting providers and potentially translate into cheaper prices for customers. For enterprises, DeepInfer could lower the cost of running DeepSeek’s models on their own infrastructure. As yet, there is no independent benchmarking of DeepInfer’s efficiency gains.

While DeepInfer applies mainly to DeepSeek’s own models, other developers could study the open-source software and adapt its techniques for their own inferencing systems.