OpenAI anticipates that frontier AI models, including its GPT family, will become more affordable in the future.
During the VB Transform 2024 event, Olivier Godement, OpenAI’s head of API Product, engaged with Carl Franzen from the media to discuss various aspects of OpenAI’s business and technology, including the significant reductions in inference costs for its models.
0:06/14:43 Are you ready for AI agents? “We launched GPT-4, the first iteration, about 15 months ago. Since then, the cost per token has dropped by 85-90%. There's no reason to believe that this trend won't continue," Godement noted while discussing the more cost-effective GPT-4o model and the industry landscape.
Godement also addressed the company’s collaboration with Apple and the dynamics of rival companies developing superior products in the marketplace.
Massive Cost-Driven Uptake for GPT-4o
When OpenAI introduced GPT-4o at its Spring Update event, it emphasized that the new model was priced at half that of GPT-4 Turbo—the company’s most robust model at the time—and was twice as fast.
Following the announcement, Godement reported a substantial increase in demand for the “world-leading” model, with users eager to transition quickly. He attributed this demand to two critical initiatives: enhancing capability and affordability.
“Users have experienced GPT-4o's multimodal features, enabling a more human-like interaction with voice. However, the efforts to reduce costs have largely gone unnoticed,” he explained.
Godement highlighted that reducing costs has allowed OpenAI to unlock new customer use cases. This means existing applications have become more cost-effective, while previously deemed too expensive options are now viable.
He expects ongoing efforts to optimize costs at both hardware and inference levels to continue driving down the expenses associated with frontier AI models—akin to trends seen with smartphones and televisions.
“We’re not focused on maximizing profit margins; our goal is to empower users to innovate and explore new use cases. Each cost optimization allows us to lower prices for our customers. I believe we have not yet reached the limit regarding intelligence and affordability,” Godement stated.
ChatGPT Enterprise Continues to Thrive
Highlighting the affordability of OpenAI models, Godement mentioned that ChatGPT Enterprise, which boasted over 600,000 users in April, remains strong, with teams across consulting, finance, marketing, and sales utilizing the platform for knowledge work. For instance, Moderna employees used ChatGPT to create a model for calculating vaccine dosages during clinical trials.
“If you look ahead a few months or years, every employee in enterprises will have a super assistant to enhance productivity and job satisfaction,” he predicted.
Godement also shared his perspective on advancements from competitors, describing a dual reaction: one of excitement about innovation in the field, and a need to strengthen customer relationships and trust. He noted that fostering strong relationships helps mitigate customer turnover, despite minor fluctuations in vendor offerings.
Lastly, he discussed the strategic reasons behind OpenAI's recent collaboration with Apple.
Ranking
