OpenAI CEO Sam Altman said one topic was top of mind at the Billionaires Summer Camp in Sun Valley: spending on AI.
In an interview with CNBC at the annual Allen & Company Sun Valley Conference, Altman said this is the first year that spending on AI has been a “very big topic” at the conference.
“Everyone is asking what they can do to reduce spending and increase value,” he said. He added that OpenAI is seeing people focus on efficiency and reaping the full benefits of investing in AI.
He said this was a key consideration when AI Lab developed its latest models, the GPT-5.6 family. New models released Thursday include the flagship Sol model, the well-rounded everyday Terra model, and the cost-effective Luna model.
Altman said GPT-5.6 Sol is “54% more token efficient for agent coding tasks,” and that OpenAI developed it with cost and speed in mind. The executive did not say what the 54% figure was compared to.
He said “every company” is now thinking about “the spend and the value they can get in exchange for AI.”
The Sun Valley Conference kicked off Tuesday, bringing together the world’s biggest technology leaders in the picturesque resort town of Sun Valley, Idaho. This year’s attendee list included Netflix co-CEO Ted Sarandos, Uber CEO Dara Khosrowshahi, Apple CEO Tim Cook, and Amazon CEO Andy Jassy.
Altman’s comments to CNBC come as companies are increasingly looking at ways to achieve higher returns from their AI spending, and several executives have publicly shared tactics to reap higher returns.
Coinbase CEO Brian Armstrong said in June that the company is experimenting with a cheaper Chinese model as the default for its engineers, and is routing prompts to the most appropriate model to avoid unnecessary burning of AI tokens.
Guillermo Rauch, CEO of cloud platform Vercel, told TechCrunch in an interview published earlier this week that companies need to start partnering with different AI labs for different parts of the AI stack. He said models from across OpenAI, Anthropic, Gemini and Chinese players should be used to get the most value.