We ran Qwen3.6-27B on $800 of consumer GPUs, day one: llama.cpp vs vLLM

Originally published at llmkube.com/blog/qwen3-6-27b-bakeoff . Cross-posted here for the dev.to audience. A Kubernetes-native bake-off on 2× RTX 5060 Ti, with reproducible manifests and a cost-per-token number neither cloud nor OSS FinOps tools will tell you. This is a runtime comparison, not a model evaluation. Both llama.cpp and vLLM serve the same Qwen3.6-27B in every cell; we're measuring how the two serving stacks differ on identical work. Where cloud APIs enter in §8, it's on cost, not cap