At the VB Transform 2024 conference, Olivier Godement, the product manager of the company's API products, had a conversation with Carl Franzen from VentureBeat, discussing various topics related to OpenAI's business and technology, including the decreasing cost of their model inference. Godement talked about the updated and more affordable GPT-4o model and industry trends, stating, "We released the first version of GPT-4 about 15 months ago. Since then, the cost per token of the model has decreased by 85-90%. There is no reason to believe that this trend will not continue." The API product manager also discussed OpenAI's partnership with Apple and their strategy for responding to competitors who launch new and more performant products in the industry. GPT-4o's cost-driven adoption at scale When OpenAI first introduced GPT-4o during their spring update event, the company quickly highlighted that the price of the new model was half that of GPT-4 Turbo, which was the most powerful model at the time, while also being twice as fast. Godement mentioned that shortly after the announcement, the company witnessed a significant adoption of this "world-leading" model, with users migrating as quickly as possible. He stated that this demand primarily stemmed from two core initiatives in research: making the product more powerful and more affordable. "In terms of functionality, people have already seen the multimodal capabilities of GPT-4o. It brings a truly human-like voice interaction experience for the first time... But what people haven't fully realized is that we have been working to make the model cheaper," he said. Godement emphasized that the latter (cost reduction) particularly helped OpenAI enable new use cases for their customers. Essentially, existing use cases became more economically efficient, and previously unconsidered use cases became feasible due to the reduced cost. He anticipated that the company would continue optimizing costs at the hardware and inference levels, further lowering the operational costs of cutting-edge AI models, similar to the situation with smartphones and televisions. "Our business is not about maximizing profit margins but enabling people to build more, try more use cases, and see which ones succeed. Every time we optimize costs, we pass on the savings to our customers. My intuition is that we are far from reaching the limit, both in terms of intelligence and cost," added Godement. ChatGPT Enterprise maintains strong momentum In addition to highlighting the affordability of OpenAI's models, Godement also pointed out that ChatGPT Enterprise had over 600,000 users in April and continued to perform strongly in various industry teams, such as consulting, finance, marketing, and sales, where it was used for knowledge work. Moderna employees used this service to create a GPT that calculated vaccine dosages for patients undergoing clinical trials. "If you zoom out and fast forward a few months or years, every employee in an enterprise will have a super assistant to help them improve work efficiency and job satisfaction," he said. Furthermore, Godement emphasized his attitude towards competitors launching improved models, stating that these developments inspired two aspects in his mind. On one hand, he was "happy" to see significant progress and innovation in the field. On the other hand, he doubled down on efforts to strengthen customer relationships and trust. He mentioned that strong customer relationships could prevent customer churn, even though he hadn't observed a significant loss of customers due to minor changes in supplier metrics.