April 3, 2011 - Last week, the company, which is responsible for maintaining and managing the ARC-AGI The Arc Prize Foundation's assessment of the OpenAI The cost estimate for the o3 "reasoning" AI model in the ARC-AGI benchmarking has been significantly revised and it now appears that o3 is not as inexpensive to run.
When OpenAI launched the o3 model last December, in collaboration with the developers of ARC-AGI, it demonstrated the model's power to solve complex problems. A few months later, however, its cost estimates have changed dramatically. the Arc Prize Foundation's latest estimate of the o3 model's best-performing configuration, o3 high, hasThe cost of solving a single ARC-AGI problem could be as high as about $30,000 (note: about $218,000 at current exchange rates), compared to previous estimates of only about $3,000 (about $21,821 at current exchange rates).
This significant upward revision of the cost estimate highlights the high cost that current state-of-the-art AI models may face for specific tasks. While OpenAI has not yet announced pricing for the o3 model or officially released the model, the Arc Prize Foundation believes that the pricing of OpenAI's most expensive current o1-pro model can be used as a reference.
"We think o1-pro is a closer comparator to the true cost of o3 because both used a similar amount of computation during testing." Mike Knoop, co-founder of the Arc Prize Foundation, said in an interview with TechCrunch, "But it's still just a reference value, and we're still marking o3 as a preview on our charts to reflect the uncertainty until official pricing is announced. "
The high cost of o3 high is not unfounded. According to the Arc Prize Foundation, o3 high uses 172 times the computational resources of the least computationally intensive o3 low configuration of the o3 model when processing ARC-AGI tasks. Such a large consumption of computational resources drives up the cost of o3 high significantly.
There have also been long-standing rumors that OpenAI plans to charge a hefty fee for high-end, customized services for enterprise customers. In early March, The Information reported that the company may plan to charge up to $20,000 per month (about Rs. 145,000 at current exchange rates) to provide specialized AI "proxy" services like software developer agents to enterprises.
While some argue that even the most expensive AI models cost far less than the salary of a human contractor or employee, an article by AI researcher Toby Ord on the X website suggests that these models may not be as efficient as one might expect. For example, o3 high required 1024 attempts to achieve optimal performance on the ARC-AGI test.
As AI technology continues to evolve and commercialization accelerates, how to effectively control costs while maintaining high performance of models will become one of the key challenges facing the industry.
