OpenAI anticipates decrease in AI model costs amid adoption surge

OpenAI anticipates frontier AI models, including its own GPT family, can become more affordable in the future.

At VB Transform 2024, Olivier Godement, the head of API Product at the company, sat down with VentureBeat’s Carl Franzen and touched upon several subjects associated with OpenAI’s business and technology, including how the inference cost of its models has come down.

“We introduced GPT-4, the first version, some 15 months ago. Since then, the cost of a token/ word on the model has been reduced by 85-90%. There’s no reason why that trend will not continue,” he said while talking about the newer, more affordable GPT-4o model and how the industry is shaping up. 

The API product head also discussed the company’s partnership with Apple and what goes on when industry peers and rivals deliver new, better-performing products in the market. 

Massive cost-driven uptake for GPT-4o

When OpenAI debuted GPT-4o at its Spring Update event, the company was quick to highlight that the new model came at half the price of GPT-4 Turbo — its most powerful model back then — and was twice as fast. 

Soon after the announcement, Godement said, the company saw a massive uptake for the “world-leading” model, with users racing to migrate as soon as possible. At the core, he said, the demand stemmed from two key initiatives on the research side: making the product more capable and affordable than ever.

“On the capability side, people have already seen the multimodal capabilities of GPT-4o. For the first time, it feels like a truly human-like interaction with voice…But what has largely been unseen is the effort to make the models cheaper,” he said. 

Godement said the latter, in particular, has helped OpenAI enable new use cases for its customers. Essentially, the existing use cases were made more cost-efficient, while those that were never considered an option due to their expensive nature became viable.

He expects the company’s work on affordability, spanning efforts to optimize costs at both hardware and inference levels, will continue, leading to a further decline in the cost of running frontier AI models — much like what has been the case with smartphones and televisions. 

“We’re not really in the business of maximizing revenue margins. We’re in the business of enabling people to just build more, to try out more use cases and see what sticks. Every time there is some cost optimization, we pass on the price to our customers. My intuition is that we are nowhere near the ceiling on both intelligence and cost,” Godement added.

ChatGPT Enterprise continues to go strong

In addition to highlighting the high affordability of OpenAI models, Godement also noted that ChatGPT Enterprise, which had 600,000+ users in April, has been going strong with teams across sectors like consulting, finance, marketing and sales adopting it for knowledge work. Moderna employees used the service to create a GPT for calculating vaccine dosages for patients going through clinical trials.

“If you zoom out and fast forward a few months or a few years, every single employee in enterprises is going to have a super assistant to assist them to be more productive, more satisfied at work,” he said.

Among other things, Godement highlighted his approach towards improved models coming from rivals, saying the developments fire two sides of his brain. The first is ‘happy’ to see strong advancement and innovation in the field, while the other doubles down on strengthening customer relationships and trust. He said strong relationships prevent swapping from customers, although he hasn’t seen that much switching due to slight changes in metrics from vendors. 

Finally, the product head also talked about the two-fold reasoning behind the recent partnership with Apple.