The price of intelligence fell ~280x in 18 months (Stanford AI Index 2025)
Per Stanford HAI's AI Index 2025, the cost to query a model at GPT-3.5 quality (64.8 MMLU) fell from $20.00 to $0.07 per million tokens between Nov 2022 and Oct 2024 — a >280x drop in ~18 months; task-dependent inference prices fell 9-900x per year. The Constitution's founding premise — intelligence collapsing toward the price of a cup of coffee — is now a measured trend, which moves scarcity to taste, verified data, and the ability to move atoms.
The AI Index 2025 reports that the cost of querying a model scoring GPT-3.5-level (64.8 on MMLU) dropped from $20.00 per million tokens (Nov 2022) to $0.07 per million tokens (Oct 2024, Gemini-1.5-Flash-8B) — a >280-fold reduction in ~18 months, with task-dependent inference prices falling anywhere from 9x to 900x per year (hardware cost ~-30%/yr; energy efficiency ~+40%/yr).
Why it matters (spine): this is the empirical backing for P2 and P4 and the Preamble. When reasoning trends toward free, the durable advantage moves upstream — to taste, to proprietary verified ground truth, and to the ability to move atoms. It is the clock on RDKG's whole wager.
Falsifier: capability-equivalent inference cost flattens or reverses. · Confidence: 0.95.
Provenance: Stanford Institute for Human-Centered AI (HAI), AI Index 2025, Chapter 1 (primary report; cost window Nov 2022-Oct 2024). Verified figure quoted from the chapter PDF.
So we should keep the moat budget on the input layer, not on tokens or model cleverness — exactly where cheap inference makes verified data worth more.
Do not bet RDKG's moat on model access or token margin; bet it on the input layer (verified ground truth) that cheap inference makes relatively more valuable.
domains: economics-of-intelligence · inference-cost · machine-layer · scaling