OpenAI highlighted its powerful reasoning capabilities in December by launching its o3 AI model. It worked with the developers of ARC-AGI to determine the model's functionality. The results were altered despite initially appearing to be highly impressive.
Initially, it was anticipated that employing o3 high to obtain a single problem would cost $3,000. Experts now project that each task might cost up to $30,000. This indicates that utilizing powerful AI models can be highly expensive. An official price for o3 has not yet been provided by OpenAI.
Arc Prize Foundation Highlights
OpenAI worked with the developers of ARC-AGI, a benchmark for testing extremely accurate AI, to indicate o3's capabilities when it introduced its o3 "reasoning" AI model in December. Despite several months, the results have been upgraded and appear somewhat less outstanding than they initially did
The Arc Prize Foundation, which manages and supervises ARC-AGI, revised its anticipated computing expenses for o3 last week. In order to fix a single ARC-AGI problem, the company first calculated that the excellent-performing o3 configuration it analyzed, o3 high, would cost about $3,000. The Arc Prize Foundation now believes the price is significantly higher, perhaps as much as $30,000 per work.
The alteration is noteworthy because it indicates how costly, at least initially, the most potent AI models available today may wind up costing for particular jobs. O3 has not yet been priced or even launched by OpenAI. However, the Arc Prize Foundation argues that OpenAI's o1-pro model pricing is a fair stand-in.
For instance, OpenAI's most costly model to date is o1-pro. Mike Knoop, one of the co-founders of the Arc Prize Foundation, stated, "We argue o1-pro is a stronger comparison of true o3 cost … because of the amount of test-time compute utilized." "However, we have maintained O3 marked as a preview on our leaderboard to indicate the uncertainty until official pricing is introduced, so this is still a proxy."
AI Pricing and Efficiency
Given how much computational power the model is stated to require, a premium price for o3 high wouldn't be outside the question. The Arc Prize Foundation states that in order to handle ARC-AGI, o3 high needed 172 times as much computational power as o3 low, the lowest-computing configuration of o3.
Furthermore, there have been long-running reports that OpenAI is considering announcing expensive plans for business clients. The corporation may be charging up to $20,000 a month for professional AI "agents," including a software development agent, based on a report published in early March by The Information.
According to some, even the most expensive OpenAI models will be far less costly than what a regular human employee or contractor would charge. However, the models might not be as efficient, as AI researcher Toby Ord claimed in a post on X. For example, in order to obtain its maximum score in ARC-AGI, o3 high needed 1,024 attempts at every activity.