Deepseek has released DeepInfer, an open-source inference framework under the Apache License 2.0, to power its V4.1-Flash model. This software manages requests and generates responses efficiently at scale by building on vLLM with custom compute kernels and distributed processing across multiple chips. The company also published deepseek-recipe, a tool that replicates the official API’s token processing so self-hosted setups match official performance. Future model code will ship as open source alongside these updates. While Deepseek’s models are already inexpensive, DeepInfer closes the final gap for running them even more cheaply. The stated goal is to help the community lower compute costs, reduce energy use, and make AI more affordable. Beyond technical efficiency, the move adds pricing pressure on Western providers like OpenAI and Anthropic. Given the ongoing AI race between China and the US, the release could be politically motivated as well. If Western AI companies miss their growth targets, that poses a risk for the US economy and beyond.
- DeepInfer allows self-hosted setups to behave identically to the official Deepseek API.
- All future Deepseek model code will be released as open source.
- The framework optimises text generation techniques for faster output.
