Plot 2 — The collapsing cost of inference ️ your turn Running a model good enough to match GPT-3.5 fell from about $20 to $0.07 per million tokens in ~18 months (≈ 280× cheaper). A basic plot is ...