The Arc Prize Foundation, which operates the synthetic intelligence (AGI) benchmark ‘ARC-AGI’, has re-evaluated the price of the O3 model of Open AI. The fee has increased significantly than the initial expectations, and expectations for the model have been barely reduced.
The Arc Prize Foundation expects that O3-High, the very best performance version of O3, will take about $ 30,000 (about 44 million won) to resolve an ARC-AGI benchmark problem.
That is the results of a major surpass of $ 3000 (about 4.4 million won) for the primary test on the time of O3 in December last yr. On the time, O3-high was applied to the O3’s rating for humans, jumping 10 times greater than that price.
In fact, open AI has not officially announced the worth of O3, and the model itself has not been officially released. The value may change.
Nevertheless, the Arc Prize Foundation saw that it will be more reasonable to make use of ‘O1-Pro’ than this cost. O1-pro can be an expensive model in open AI, however it is cheaper than O3.
It is because the high price of O3-High is big. Based on the Arc Prize Foundation, O3-Hi used 172 times more operations than the bottom operation setting, O3-Low.
For that reason, there may be a possibility that open AI will introduce expensive plans for corporate customers. The agent, which has a Ph.D. performance, is thought to have set a fee for $ 240,000 per 30 days and annual salary.
Some argue that even the costliest models of open AI could be cheaper than labor costs.
Nevertheless, Toby Old, an AI researcher, identified that the AI ​​model may not necessarily be more efficient than humans.
He said that O3-Hi has made 1024 attempts for every work to record the very best rating in ARC-AGI, and has written 137 pages of text of about 43 million words per attempt. This is similar as writing a Britannicer Encyclopedia, including a complete of 44 million words per work.
“It costs about $ 30,000 for O3 to resolve each problem,” OD said. “This is an easy puzzle that my 10 -year -old son can solve in 4 minutes.”
Alternatively, Open AI is preparing to launch the GPT-5, which integrates the O3 and the present non-theory model.
By Park Chan, reporter cpark@aitimes.com