DeepSeek open-sources the engine for running AI models

DeepSeek has open-sourced DeepInfer, the software it uses to run its V4.1-Flash model at scale. This is not a new chatbot or model: it is the layer that takes a trained model and turns a flood of user requests into answers, deciding how work is grouped and spread across expensive AI chips. By making that code available, DeepSeek gives companies that run their own models a starting point for extracting more output from the same hardware, which could lower the ongoing cost of AI services. The release is narrower than a ready-made cloud product—currently built for one DeepSeek model and specific chips—so savings will depend on a company’s technical setup and ability to operate it.