Beijing's AGI Bar Offers Free DeepSeek V4 Flash Access via Local Nvidia Workstation

AGI Bar, a programmer-focused venue in Beijing, has launched a service offering customers free, unlimited access to the DeepSeek V4 Flash model. The service is powered by an on-site Nvidia DGX Spark AI workstation, allowing customers to perform local inference by connecting to the bar's Wi-Fi. This initiative represents a novel integration of local AI hardware deployment with physical retail and social spaces, showing how AI compute can be offered as a customer amenity. It highlights the trend of declining inference costs and the potential for shared workspaces, cafes, and bars to offer localized AI services. Customers receive a Base URL and API Key to integrate the local DeepSeek V4 Flash model directly into development tools like Cursor, Cherry Studio, and Claude Code. The DeepSeek V4 Flash model is a Mixture-of-Experts (MoE) model featuring 284 billion total parameters, with 13 billion activated parameters.