Physical Address

304 North Cardinal St.
Dorchester Center, MA 02124

OpenAI’s o3 model might be costlier to run than originally estimated


When Openai opened it O3 “Justification” AI model In December, the company AGI’s creators, a benchmark designed to test the high-handed AI, a benchmark designed to demonstrate O3’s capabilities. Months later, the results are revised and now they seem slightly less impressive than they do.

Last week, the foundation of the ARC Prize, which protects the ARC-AGE and renovated the estimated calculation costs for the foundation O3. The organization first ended the best-playing configuration of O3, the height of O3 to solve a single Arc-Aggular problem. Now the ARC Award Foundation thinks that the value is higher – perhaps $ 30,000 to a position.

Inspection is noteworthy, because today’s most advanced AI models of AI shows the end of the end for certain tasks at least early. Openai still has O3 price or leave it. But the arc reward believes in the Foundation Openai O1-PRO model A reasonable proxy of valuation.

For the context, O1-Pro is the most expensive model so far.

“We believe that O1-Pro is a better comparison of TRUE-O3 value … Test time calculation amount,” Mike Knoop, “Mike Knoop,” Mike Knoop, “Mike Knoop,” said Mike Knoop, “said. “However, this is still a proxy and labeled in our leading panel to reflect uncertainty until the official price is announced.”

The high price for the high price O3 will not be left out of the question, taking into account the amount of reporting sources used by the model. According to the ARC Prize Foundation, more than 172x computing than O3, the calculation configuration than O3 to combat ARC-AGI.

In addition, rumors fled for a while due to expensive plans, Openai thinks about the application for enterprise customers. Information at the beginning of March declare The company plans to pay up to $ 20,000 per month for a program developer agent, a $ 20,000 per month for AI “Agents”.

Some can even claim that Openai’s Prtiest models will be good at a good price with a typical human contractor or employee. AI Researcher Toby Ord pointed at an article in XModels may not be effective. For example, in the Arc-Aggi to collect the best points at the high level of O3, you need to try 1,024 in each position.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *