Finance Teams Confront AI Bills as Advanced Models Use 14 Times More Tokens
Updated
Updated · The New York Times · Sep 9
Finance Teams Confront AI Bills as Advanced Models Use 14 Times More Tokens
3 articles · Updated · The New York Times · Sep 9
Summary
Finance departments are starting to flag AI spending as a major cost center after companies pushed broad employee adoption without closely tracking token use.
14 times more tokens were consumed by a newer OpenAI model than an earlier ChatGPT version for the same road-trip prompt, illustrating how more capable systems sharply raise compute bills.
10 times higher per-token costs can also apply within a single provider—Anthropic’s Mythos costs over 10 times more than a cheaper in-house model because advanced systems need newer chips and more energy.
Most of that reasoning work is invisible to users, while many chatbot plans still offer near-unlimited access at subsidized prices, masking the true cost to businesses and consumers.
That pricing pressure is becoming more important as OpenAI and Anthropic prepare for IPOs, raising questions about margins, future user charges and whether demand will hold once subsidies fade.