The Daily Hallucination

SAN FRANCISCO — OpenAI on Tuesday cut the price of its flagship models roughly in half with the release of GPT-6 Sol and GPT-6 Luna, the latter priced at 10 cents per million input tokens, prompting finance departments across the technology industry to briefly believe their AI bills were going to go down.

"For about 40 minutes it was the best day of my career," said Harriet Bloemfontein, vice president of finance at a Denver logistics software company. "Then an engineer posted in Slack that we could now afford to have the agent check its own work, and then check that, and then write a summary of the checks."

According to a survey of 212 engineering leaders by the procurement firm Ledgerline, 91 percent said the price cut would allow their teams to "do more with the same budget," and 91 percent said "the same budget" meant all of it.

"This is the oldest story in economics," said Tobias Nakamura-Ferreira, an economist who studies cloud spending. "You make coal more efficient, people burn more coal. You make tokens cheaper, an engineer builds an agent whose only job is to ask a second agent whether the first agent is sure."

At one Seattle company, the platform team responded to the announcement by raising the default number of retries from three to nine and by giving every pull request its own dedicated reviewer model, which reviews the pull request and also, on its own initiative, the reviewer.

"Our costs per token fell 50 percent this morning," said a staff engineer there. "Our tokens per anything went up by 100 percent this afternoon. So I would say we broke even, which is the first time we have broken anything even."

At press time, Bloemfontein had forwarded the price announcement to her CEO with the subject line "good news (withdrawn)."