Open Source LLMs Are Crushing Closed-Source Models on Cost — The Numbers Don't Lie The Raw Price Gap Nobody Wants to Admit Closed-source APIs still charge premium rates that scale directly with volume. GPT-4 Turbo lists at $.01 per 1,000 input tokens and $.03 per 1,000 output tokens. In contrast, self-hosted Llama-3-70B on a single A100 node drops that figure to roughly $.0004 per 1,000 tokens...