When Openai presented the O3 AI model “Collection” in December, the company worked with the creators of Arc-Agi, a reference point designed to try extremely capable AI to present O3’s capabilities. Months later, the results have been revised and now they look slightly less impressive than initially.
Last week, the ARC Awards Foundation, which maintains and manages the ARC-AGI, updated its computing costs for O3. The organization initially estimates that the O3-tested O3-tested performance costs about $ 3,000 to solve a single ARC-AGI problem. Now the ARC Awards Foundation believes that costs are much higher – Probably about $ 30,000 per job.
The review is remarkable because it shows how expensive the most sophisticated AI models of today may end up being for certain tasks, at least early. Openai has not yet honored O3 – or will release it, even. But the ARC Awards Foundation believes that OPENAI’s O1 Model pricing is a reasonable proxy.
For the frame, O1-PRO is the most expensive model of Openai to date.
“We believe that O1-PRO is a closer comparison of true O3 costs … due to the amount of testing time used,” said Mike Knoop, one of the co-founders of the Arc Prize Foundation. “But this is still a proxy and we have kept the O3 with the preview of our leaderboard to reflect uncertainty until the official pricing is announced.”
A high price for O3 HIGH would not be out of question, given the amount of computational resources used by the model. According to the ARC Awards Foundation, the O3 High used 172X more computers than O3 Low, the lower O3 computation to address the ARC-AGI.
In addition, rumors have been flying for some time about expensive Openai designs are thinking of introducing business customers. In early March, information said the company could plan to charge up to $ 20,000 a month for specialized “agents”, such as a software developer.
Some may argue that even the most expensive models of Openai will cost well under what a typical contractor or employee would have. But as a researcher AI Toby Ord pointed to a post in xThe models may not be as effective. For example, the O3 High needed 1,024 attempts at every work on Arc-AGI to achieve his best score.
