DeepSeek opens up AI software aimed at reducing running costs
DeepInfer currently supports V4.1-Flash, with more models to follow, but its efficiency gains have not been independently benchmarked.
DeepSeek has made DeepInfer, the software behind its large-scale AI operations, available as open source, potentially helping businesses run its models at lower cost. The Chinese company said on X that sharing the software was intended to reduce computing expenses and energy consumption.
The software currently works with DeepSeek’s V4.1-Flash model. DeepSeek said it would add support for future systems. Cloud operators and hosting providers can use the software to apply the company’s methods when serving its models.
DeepInfer combines techniques including distributed inference, which divides AI workloads among hardware resources. DeepSeek has adapted these methods for its own models to increase output and improve chip utilisation. Inference refers to running AI models.
Thibault Jaigu, chief executive officer of Requesty, an AI gateway metrics and router startup, said the software made hosting DeepSeek models more accessible. However, operators would still need computing capacity and technical expertise.
More efficient operation could improve hosting providers’ margins and potentially lower customer prices, while businesses could spend less running models on their own infrastructure. However, DeepInfer’s efficiency gains have yet to undergo independent benchmarking. Developers working on other models could also examine the code and adapt its methods.
