DeepSeek open-sources DeepInfer framework to slash AI operating costs

Hangzhou-based artificial intelligence startup DeepSeek has open-sourced DeepInfer, the internal software framework used to run its V4.1-Flash model, making the tool freely available to developers on GitHub.

DeepSeek said in a statement on X that it published its production stack to help the wider technical community lower compute costs, reduce energy consumption, and make AI deployment more affordable. The company added that future models will be integrated into the framework as they are released.

Inference—the computing process where a trained model processes prompts and generates outputs—has become the largest ongoing expense in corporate AI budgets. As businesses shift towards multi-step automated agents that perform complex tasks, the computing power required to run these systems has climbed steeply.

DeepSeek's model runs at a small fraction of the cost charged by leading American competitors, placing pressure on silicon valley firms to justify their higher price points.

By incorporating techniques such as speculative decoding, DeepInfer is designed to further reduce running costs for organisations hosting the model. The open-source release comes as corporate technology departments face increasing pressure to curb spiralling AI expenses.

The decision to make the software freely available also follows action by the Trump administration to ban US federal civilian agencies, including the Department of State, from using DeepSeek's tools over national security concerns. However, the federal directive does not prohibit private companies or commercial organisations from deploying the software, leaving American firms free to adopt the underlying framework to lower their technical overheads.