DeepSeek V4 Flash 0731 turns dual DGX Sparks into an AI workhorse

DeepSeek's V4 Flash 0731 (DSV4F 0731), running locally on dual NVIDIA DGX Sparks, is being called a game-changer by an early adopter running a small company. The user, who previously used a Q3.6 27B full FP8 on dual RTX 3090s, describes the upgrade as “a whole new level.” In daily use, DSV4F 0731 powers an Hermes agent for everyday tasks, handles two-hour coding sessions with OpenCode without slowing down, and integrates with Paperless NGX for document retrieval and DOCX automation. The user also wrote a custom skill to automate paperwork completion, calling it “easy peasy.”

The real kicker: the user says client tickets are now solved by copy-pasting from the ticketing system, with the AI generating responses and closing out issues. That direct time-to-money workflow convinced them to order a second pair of DGX Sparks, despite the hardware cost. They noted that on-device models like DeepSeek's get better over time without additional fees, making the investment feel justified. This anecdote highlights a broader shift: capable local AI models running on dedicated hardware are replacing cloud APIs for many SMB workloads, offering privacy, control, and ever-improving performance. The user is now planning a ticket system integration over the weekend, hinting that even more automation is on the way.