When Openai discovered his O3 model “reasoning” in December, the company partnership with the creators of ARC-AGI, a landmark created to prove he very capable, to show O3 skills. Months later, the results have been revised, and they now look a little less impressive than originally.
Last week, the Arc Arc Foundation, which maintains and administers ARC-AGI, updated its approximate O3 computing costs. The organization initially estimated that configuration with the best O3 tested O3 performance costs about $ 3,000 to solve a single ARC-AGI problem. Now, the ARC Price Foundation thinks the cost is much higher – perhaps about $ 30,000 for duty.
The review is evident because it illustrates how expensive he can finish the most sophisticated models of he for certain tasks, at least early. Openai has not yet appreciated O3 – or let go of it, even. But the ARC Awards Foundation believes that the Openai O1-Pro Price is a reasonable representative.
For the context, O1-Pro is the most expensive model of Openai so far.
“We believe that O1-Pro is a narrower comparison of the real cost O3 (…) due to the amount of test time calculated,” said Mike Knoop, one of the co-founders of the ARC Arc-Tycrunch Foundation. “But this is still a representative, and we have kept O3 labeled as a survey in our table to reflect uncertainty until the official price is announced.”
A high price for O3 High would not be out of the question, given the amount of computing resources that the model reported to use. According to the ARC Price Foundation, O3 High used 172x more computing than low O3, lower O3 calculation configuration to handle ARC-AGI.
Moreover, rumors have been flying for a long time about the costly plans Openai is thinking of presenting to enterprise clients. In early March, information reported that the company could plan to upload up to $ 20,000 a month for it as a software developer agent.
Some may argue that even the most precious Openai models will cost well under what would command a typical human contractor or personnel. But as Ai Toby Ord’s researcher points out in a post on X, the models may not be as effective. For example, O3 High needed 1,024 effort for each task at the ARC-AGI to achieve its best result.
