DeepSeek seeks ‘total victory’ over closed-source AI with new open-source tool

DeepSeek has fully open-sourced DeepInfer, an internally developed inference framework built to run its DeepSeek-V4.1-Flash model efficiently on Nvidia's SM100 Blackwell GPUs and Huawei hardware.

• The release, made on Tuesday, is framed by a senior company researcher as part of a wider contest between open-source and closed-source AI, described as a battle between "light and darkness".
• DeepInfer is tailored specifically to support DeepSeek-V4.1-Flash, and the company says it was developed in-house rather than adapted from outside work.
• By publishing the framework, DeepSeek is handing outside developers the same tooling it uses to serve its own model, which the company presents as a way to lower the cost of running AI.