Altman Claims Token Pricing Will Become Obsolete: GPT-5.5 More Expensive but Usage Drops, Customers Only Care About Task Completion

According to monitoring by Beating, tech commentator Ben Thompson published a joint interview with OpenAI CEO Altman and AWS CEO Matt Garman on Stratechery, focusing on their recently announced Bedrock Managed Agents. In the interview, Altman discussed pricing, asserting that token-based pricing is not sustainable in the long run. He cited GPT-5.5 as an example: the price per token is significantly higher than that of GPT-5.4, but the number of tokens used for the same task is much lower. Users are not concerned with the number of tokens; they only care about whether the task is completed and the total cost. He positioned OpenAI as an ‘intelligence factory’ rather than a token factory, believing that customers want to spend the least amount of money to acquire the most intelligence, regardless of the underlying model or the number of tokens used. Altman also mentioned that currently, there are far more customers requesting increased capacity than those looking to negotiate prices. He compared AI to utilities: no matter how cheap water is, you won’t drink excessively, but intelligence is different; ‘as long as the price is low enough, I will keep using more, and no other utility works this way.’ Garman added that the price of computing power has decreased by numerous orders of magnitude over the past 30 years, yet the amount of computing power sold today is greater than ever. Thompson noted that Altman confidently stated in 2023 that AI would threaten Google Search and asked for his thoughts in hindsight. Altman said the outcome was better than expected, but in unexpected ways: it was initially thought that AI would change how people search for information online and directly impact search engines; in reality, OpenAI’s success mainly stemmed from ChatGPT as an independent consumer product and the Codex API, and search itself has not been disrupted. He described ChatGPT as ‘the first truly large-scale consumer product since Facebook,’ while also acknowledging that Google is still underestimated in many respects.