Long-context LLM serving: the real tradeoffs in memory, latency, cost, and accuracy

What long-context LLM requests really cost: KV cache math, latency, per-session pricing, and when RAG still wins.