1.6 trillion parameters. 49B active per token. Too large for a single node. We deployed DeepSeek-V4-Pro on Hyperstack using multi-node Kubernetes - 16 NVIDIA H100s across two worker nodes, hybrid Data + Expert Parallelism, and a 960 GB FP4+FP8 checkpoint loaded from local NVMe. In this tutorial: → Multi-node Kubernetes cluster on Hyperstack (2x 8x NVIDIA H100-80G PCIe-NVLink) → LeaderWorkerSet API for coordinated 2-node inference → vLLM with hybrid DEP topology and MTP speculative decoding → 1M token context window with three reasoning tiers → Long-horizon autonomous code refactoring with self-correction → Plugging into Claude Code, OpenClaw, and OpenCode as a local backend 80.6 on SWE-Bench Verified. 93.5 on LiveCodeBench v6. Full tutorial on the blog: bit.ly/4f1jamb #DeepSeek #AgenticAI