Talking AI — Inference cost: the price of AI dropped 280-fold in two years

Inference cost is what each answer produced by a deployed AI actually costs — the operating bill, distinct from the project budget. According to Stanford, that price has collapsed: GPT-3.5-level performance cost $20 per million tokens in November 2022, versus $0.07 in October 2024 — 280 times less in two years. The consequence: an operating budget computed at today's prices will be wrong within a year, and a multi-year fixed-price contract hands the entire decline to the vendor rather than the client. Any SMB or mid-sized company contracting an AI project should require an annual price-review clause indexed to model costs, and measure the unit cost per task processed from month one.

Source: Stanford HAI, "Artificial Intelligence Index Report 2025," April 2025.