DeepSeek Fully Opens DeepInfer Framework Used for V4.1 Flash

DeepSeek has made DeepInfer, its internal inference framework, fully open source. The framework is already used to run a large-scale inference service for DeepSeek V4.1 Flash, putting the software used to serve that model in the open-source project.

DeepSeek said the service delivers high throughput and low latency. It described this as an operating service built on DeepInfer, rather than a future deployment. The stated use links the framework directly to V4.1 Flash’s current inference workload.

The company plans to publish inference code for new models through DeepInfer. It also plans to continue adding later optimizations to the project. DeepSeek’s stated plan puts both model-specific inference code and follow-up optimization work in the same framework as it introduces new models.

V4.1 Flash is the model identified as running on DeepInfer. DeepSeek’s future plan centers on continuing to publish inference code and optimization updates through the framework, tying its current use to ongoing work on later models.