The bill covered 603 billion tokens across 7.6 million requests from 100 Codex instances running GPT-5.5. Disabling Fast Mode would cut the cost to $300,000, but the figure reveals the true economics of autonomous AI development.