DeepSeek releases DeepInfer under Apache 2.0 for offline V4.1 Flash inference on NVIDIA SM100 and Huawei Ascend 950, alongside deepseek-recipe API codecs; a complete serving stack requires additional infrastructure.

DeepSeek’s own DeepInfer V4.1 Flash decode benchmark chart, showing its reported output-throughput versus per-user-speed tradeoff; not an independent performance test.

Published 20 Oct 2026 · 10:37 UTC

DeepSeek ↗ · Source dated 20 Oct 2026

DeepSeek releases DeepInfer under Apache 2.0 for offline V4.1 Flash inference on NVIDIA SM100 and Huawei Ascend 950, alongside deepseek-recipe API codecs; a complete serving stack requires additional infrastructure. · ournews.ai