❓ Is Altman’s law a measured universal price series or a named observation about the cost of a fixed capability?
On 9 February 2025, Sam Altman wrote: “The cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use.” His evidence example: the token cost of GPT-4 in early 2023 compared with GPT-4o in mid-2024, which he described as about 150× lower. A token is a piece of text that a model reads or writes and an API may bill for. The post does not provide a table of transaction prices across models or a method for deriving a quality-adjusted index. [1]
The live Future Forge Altman tile deliberately labels the ~10× annual decline a named observation. It sets the GPT-4-era comparison at 150 and the GPT-4o-era point at 1, with an index unit (“price compared with mid-2024”). Those are two points representing Altman’s pair, not a fresh dataset for 2025–26 and not proof that every enterprise workload got 150× cheaper. [3]
Holding “intelligence level” constant is itself a judgment: different models differ on different tasks. His post uses a broad capability description and a token price. It does not assert that the dollar cost of a completed, quality-checked task falls at exactly 10× each year.
Classification: named exponential price observation over 2023–24 with a hypothesized 10× annual rate. Possible drivers include more efficient models, improved hardware, and competition, but Altman’s post does not decompose their contributions. It is a claim to test against other meters, not an audited index.