DeepSeek open-sources DeepInfer, the production inference engine behind V4.1 Flash

DeepSeek open-sourced DeepInfer on October 20, 2026, the inference framework it says powers its large scale production services, including DeepSeek-V4.1-Flash, released under the Apache 2.0 license. The release includes high-performance kernels, distributed inference, speculative decoding and structured outputs, plus deepseek-recipe, the production token encoding and decoding component aligned with the official DeepSeek API so community deployments behave like the API. It supports Nvidia SM100 GPUs and Huawei Ascend 950 chips, and its primary target is offline inference. "We're sharing our production stack to help the community lower compute costs, reduce energy use and make AI more affordable," DeepSeek said in its announcement post.

The release omits pieces of DeepSeek's own serving stack: the KV-cache system, disaggregated serving and load balancing are not included, and DeepSeek's own efficient reference setup is 64 GPUs for prefill plus 64 for decode, so community deployments will not yet match DeepSeek's own serving costs. The release came a day after the Department of Homeland Security order barring federal civilian agencies from DeepSeek products took public effect, an order already listed on this page.

Sources: PYMNTS; TechTarget; The Diary of AI