DeepSeek Open-Sources DeepInfer to Cut AI Inference Costs for Enterprises

China's DeepSeek on Tuesday released DeepInfer, an open-source software system for running its AI models at scale, giving enterprises, cloud providers and model-hosting firms access to its cost-reduction methods. The Hangzhou-based company announced the release on X, saying it aims to help developers lower computing costs and energy consumption. No independent performance tests yet verify the actual gains. DeepInfer currently supports DeepSeek's V4.1-Flash model, with support planned for future systems. The release targets the inference stage—processing user requests after a model is trained—where software choice can sharply affect cost per request. The system combines distributed inference, which spreads workloads across hardware, with techniques to boost chip utilization. AI companies rarely disclose full production optimization methods, and the release lets cloud and hosting providers deploy the technology themselves. Lower inference costs could improve hosting margins or pass through to customer pricing, though actual benefits remain untested. Developers outside DeepSeek's ecosystem may also adapt the code for their own systems.