OpenAI partnered with ARC-AGI creators to showcase o3’s capabilities when it unveiled its “reasoning AI” model in December. ARC is a benchmark that tests highly capable AI. The results are now revised and look less impressive than initially.
The Arc Prize Foundation (which maintains and administers ARC-AGI) updated its approximate computing cost for o3 last week. The organization estimated that the highest-performing configuration of O3 it tested, called o3 high cost around $3,000 for ARC-AGI problems. Now the Arc Prize Foundation believes that the cost of solving a single ARC-AGI problem is much higher. Around $30,000 per job
This revision is noteworthy because it shows how expensive the most sophisticated AI models today may end up costing for certain tasks. OpenAI has not yet priced o3 – or released it. The Arc Prize Foundation, however, believes that OpenAI’s pricing for the o1-pro is a reasonable substitute.
To put things in perspective, o1 is OpenAI’s priciest model to date.
Mike Knoop is one of the founders of the Arc Prize Foundation. He told TechCrunch that o1-pro was a better comparison to the true cost of o3 due to the amount of test-time computation used. “But this is just a proxy. We’ve labeled o3 as preview on our Leaderboard to reflect the uncertainties until official pricing is announced.” According to the Arc Prize Foundation o3 high used 172x as much computing power than o3 low – the lowest configuration of o3 – to tackle ARC AGI.
Furthermore, rumors about OpenAI’s expensive plans for enterprise customers have been circulating for a long time. The Information reported in early March that the company was planning to charge as much as $20,000 per month to “agents” who are specialized AI, such a software developer.
Some may argue that OpenAI’s most expensive models will cost less than what a typical contractor or staffer would charge. AI researcher Toby Ord says: In a post by Xit was pointed out that the models might not be as efficient. For example, o3 required 1,024 attempts to complete each task within ARC-AGI in order to achieve its highest score.
Kyle Wiggers, TechCrunch AI Editor. His writings have appeared in VentureBeat, Digital Trends and a variety of gadget blogs, including Android Police and Android Authority, Droid-Life and XDA-Developers. He lives in Manhattan, with his music therapist partner.
View Bio
