DeepSeek Open-Sources DeepInfer, the Inference Engine Behind Its API

DeepSeek has open-sourced DeepInfer, the in-house inference framework that powers its API in production — the engine behind DeepSeek-V4.1-Flash — under the Apache 2.0 license.

What's in the release

DeepInfer includes DeepSeek's optimized computing kernels and communication libraries, alongside mechanisms for distributing inference across multiple chips:

DeepInfer — the core inference framework
FlashMLA
DeepGEMM
DeepEP

Why it matters

DeepSeek had previously indicated that a full release of its inference engine was not practical. With this release, that engine is now fully open, with the company framing the move as an effort to make AI more affordable.

The multi-chip distribution is notable for developers: inference is not locked to a single hardware vendor. Coverage of the release has highlighted that the framework extends to non-Nvidia silicon, with Huawei receiving equal billing alongside Nvidia in the stack.

For teams running local and on-prem inference, an open, Apache-2.0-licensed inference framework — with the exact kernels a frontier lab uses in production — is a significant reference point to build on.

Source: DeepSeek